Datasets:
Dataset Viewer
id
stringlengths 24
37
| image
imagewidth (px) 270
720
| conversations
listlengths 2
2
|
|---|---|---|
aitw_17021440438302937530_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 2 bedroom apartment rent for in Chicago?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"HOME\"}"
}
] |
|
aitw_17021440438302937530_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 2 bedroom apartment rent for in Chicago?\n\nPrevious actions:\nStep 0: Press [HOME]\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"HOME\"}"
}
] |
|
aitw_17021440438302937530_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 2 bedroom apartment rent for in Chicago?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Press [HOME]\n\nCurrent screenshot is labeled with 10 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.61, 0.85]}}"
}
] |
|
aitw_17021440438302937530_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 2 bedroom apartment rent for in Chicago?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Press [HOME]\nStep 2: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 23 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"2 bedroom apartment rent for in Chicago\"}"
}
] |
|
aitw_17021440438302937530_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 2 bedroom apartment rent for in Chicago?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Press [HOME]\nStep 2: Tap [somewhere on the screen]\nStep 3: Type [2 bedroom apartment rent for in Chicago]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"ENTER\"}"
}
] |
|
aitw_17021440438302937530_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 2 bedroom apartment rent for in Chicago?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Press [HOME]\nStep 2: Tap [somewhere on the screen]\nStep 3: Type [2 bedroom apartment rent for in Chicago]\nStep 4: Press [ENTER]\n\nCurrent screenshot is labeled with 16 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 14}}"
}
] |
|
aitw_17021440438302937530_6 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 2 bedroom apartment rent for in Chicago?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Press [HOME]\nStep 2: Tap [somewhere on the screen]\nStep 3: Type [2 bedroom apartment rent for in Chicago]\nStep 4: Press [ENTER]\nStep 5: Tap [Chicago, IL - River West - Apartments starting at... Now leasing Studio &3 Bedroom Apartments in Chicago's River West Neighborhood. Luxury Apartments Located in..]\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 6) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_4002393414721824513_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Find the nearest electronics store that's open\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 10 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"HOME\"}"
}
] |
|
aitw_4002393414721824513_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Find the nearest electronics store that's open\n\nPrevious actions:\nStep 0: Press [HOME]\n\nCurrent screenshot is labeled with 9 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.61, 0.85]}}"
}
] |
|
aitw_4002393414721824513_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Find the nearest electronics store that's open\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 23 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"Find the nearest electronics store that's open\"}"
}
] |
|
aitw_4002393414721824513_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Find the nearest electronics store that's open\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\nStep 2: Type [Find the nearest electronics store that's open]\n\nCurrent screenshot is labeled with 22 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"ENTER\"}"
}
] |
|
aitw_4002393414721824513_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Find the nearest electronics store that's open\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\nStep 2: Type [Find the nearest electronics store that's open]\nStep 3: Press [ENTER]\n\nCurrent screenshot is labeled with 18 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_12106880758183517332_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Install the Google app\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"HOME\"}"
}
] |
|
aitw_12106880758183517332_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Install the Google app\n\nPrevious actions:\nStep 0: Press [HOME]\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SWIPE\": \"UP\"}"
}
] |
|
aitw_12106880758183517332_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Install the Google app\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\n\nCurrent screenshot is labeled with 32 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 7}}"
}
] |
|
aitw_12106880758183517332_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Install the Google app\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\nStep 2: Tap [PLAY Icon]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 4}}"
}
] |
|
aitw_12106880758183517332_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Install the Google app\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\nStep 2: Tap [PLAY Icon]\nStep 3: Tap [Search for ...]\n\nCurrent screenshot is labeled with 6 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"google app\"}"
}
] |
|
aitw_12106880758183517332_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Install the Google app\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\nStep 2: Tap [PLAY Icon]\nStep 3: Tap [Search for ...]\nStep 4: Type [google app]\n\nCurrent screenshot is labeled with 15 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 10}}"
}
] |
|
aitw_12106880758183517332_6 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Install the Google app\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\nStep 2: Tap [PLAY Icon]\nStep 3: Tap [Search for ...]\nStep 4: Type [google app]\nStep 5: Tap [google app]\n\nCurrent screenshot is labeled with 16 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 6) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 10}}"
}
] |
|
aitw_12106880758183517332_7 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Install the Google app\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\nStep 2: Tap [PLAY Icon]\nStep 3: Tap [Search for ...]\nStep 4: Type [google app]\nStep 5: Tap [google app]\nStep 6: Tap [Google Google L.. Tools Browsers • • Installed B]\n\nCurrent screenshot is labeled with 14 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 7) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_8714679281901806001_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 3 bedroom apartment rent for in Washington DC?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 9 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"HOME\"}"
}
] |
|
aitw_8714679281901806001_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 3 bedroom apartment rent for in Washington DC?\n\nPrevious actions:\nStep 0: Press [HOME]\n\nCurrent screenshot is labeled with 8 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.65, 0.85]}}"
}
] |
|
aitw_8714679281901806001_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 3 bedroom apartment rent for in Washington DC?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 24 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"3 bedroom apartment rent for in Washington DC?\"}"
}
] |
|
aitw_8714679281901806001_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 3 bedroom apartment rent for in Washington DC?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\nStep 2: Type [3 bedroom apartment rent for in Washington DC?]\n\nCurrent screenshot is labeled with 18 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"ENTER\"}"
}
] |
|
aitw_8714679281901806001_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 3 bedroom apartment rent for in Washington DC?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\nStep 2: Type [3 bedroom apartment rent for in Washington DC?]\nStep 3: Press [ENTER]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SWIPE\": \"UP\"}"
}
] |
|
aitw_8714679281901806001_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 3 bedroom apartment rent for in Washington DC?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\nStep 2: Type [3 bedroom apartment rent for in Washington DC?]\nStep 3: Press [ENTER]\nStep 4: Swipe [UP]\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 10}}"
}
] |
|
aitw_8714679281901806001_6 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 3 bedroom apartment rent for in Washington DC?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\nStep 2: Type [3 bedroom apartment rent for in Washington DC?]\nStep 3: Press [ENTER]\nStep 4: Swipe [UP]\nStep 5: Tap [Studios, 1-3 Bedroom Apts Washington DC... Market House Apartments Is ldeally * Discover]\n\nCurrent screenshot is labeled with 15 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 6) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 12}}"
}
] |
|
aitw_8714679281901806001_7 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How much does a 3 bedroom apartment rent for in Washington DC?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [somewhere on the screen]\nStep 2: Type [3 bedroom apartment rent for in Washington DC?]\nStep 3: Press [ENTER]\nStep 4: Swipe [UP]\nStep 5: Tap [Studios, 1-3 Bedroom Apts Washington DC... Market House Apartments Is ldeally * Discover]\nStep 6: Tap [IAgree]\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 7) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_8279218588340882978_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What is the capital of Japan?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 13 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 12}}"
}
] |
|
aitw_8279218588340882978_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What is the capital of Japan?\n\nPrevious actions:\nStep 0: Tap [9]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 16}}"
}
] |
|
aitw_8279218588340882978_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What is the capital of Japan?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [9]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 16}}"
}
] |
|
aitw_8279218588340882978_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What is the capital of Japan?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [9]\nStep 2: Tap [9]\n\nCurrent screenshot is labeled with 37 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 15}}"
}
] |
|
aitw_8279218588340882978_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What is the capital of Japan?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [9]\nStep 2: Tap [9]\nStep 3: Tap [How big is the moon?]\n\nCurrent screenshot is labeled with 25 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 2}}"
}
] |
|
aitw_8279218588340882978_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What is the capital of Japan?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [9]\nStep 2: Tap [9]\nStep 3: Tap [How big is the moon?]\nStep 4: Tap [X Icon]\n\nCurrent screenshot is labeled with 49 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"capital of Japan?\"}"
}
] |
|
aitw_8279218588340882978_6 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What is the capital of Japan?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [9]\nStep 2: Tap [9]\nStep 3: Tap [How big is the moon?]\nStep 4: Tap [X Icon]\nStep 5: Type [capital of Japan?]\n\nCurrent screenshot is labeled with 25 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 6) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"ENTER\"}"
}
] |
|
aitw_8279218588340882978_7 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What is the capital of Japan?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [9]\nStep 2: Tap [9]\nStep 3: Tap [How big is the moon?]\nStep 4: Tap [X Icon]\nStep 5: Type [capital of Japan?]\nStep 6: Press [ENTER]\n\nCurrent screenshot is labeled with 32 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 7) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_5702016030630049033_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How do I get to the nearest Nordstrom?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 15 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"HOME\"}"
}
] |
|
aitw_5702016030630049033_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How do I get to the nearest Nordstrom?\n\nPrevious actions:\nStep 0: Press [HOME]\n\nCurrent screenshot is labeled with 14 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 13}}"
}
] |
|
aitw_5702016030630049033_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How do I get to the nearest Nordstrom?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [9]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 16}}"
}
] |
|
aitw_5702016030630049033_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How do I get to the nearest Nordstrom?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [9]\nStep 2: Tap [9]\n\nCurrent screenshot is labeled with 53 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 21}}"
}
] |
|
aitw_5702016030630049033_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How do I get to the nearest Nordstrom?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [9]\nStep 2: Tap [9]\nStep 3: Tap [flights from Buenos aires to Helsinki]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 2}}"
}
] |
|
aitw_5702016030630049033_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How do I get to the nearest Nordstrom?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [9]\nStep 2: Tap [9]\nStep 3: Tap [flights from Buenos aires to Helsinki]\nStep 4: Tap [X Icon]\n\nCurrent screenshot is labeled with 55 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"How do I get to the nearest Nordstrom?\"}"
}
] |
|
aitw_5702016030630049033_6 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How do I get to the nearest Nordstrom?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [9]\nStep 2: Tap [9]\nStep 3: Tap [flights from Buenos aires to Helsinki]\nStep 4: Tap [X Icon]\nStep 5: Type [How do I get to the nearest Nordstrom?]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 6) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"ENTER\"}"
}
] |
|
aitw_5702016030630049033_7 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How do I get to the nearest Nordstrom?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Tap [9]\nStep 2: Tap [9]\nStep 3: Tap [flights from Buenos aires to Helsinki]\nStep 4: Tap [X Icon]\nStep 5: Type [How do I get to the nearest Nordstrom?]\nStep 6: Press [ENTER]\n\nCurrent screenshot is labeled with 46 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 7) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_6915940817252476665_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Check the settings for the Spotify app\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 14 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 13}}"
}
] |
|
aitw_6915940817252476665_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Check the settings for the Spotify app\n\nPrevious actions:\nStep 0: Tap [9]\n\nCurrent screenshot is labeled with 35 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.59, 0.81]}}"
}
] |
|
aitw_6915940817252476665_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Check the settings for the Spotify app\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 11}}"
}
] |
|
aitw_6915940817252476665_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Check the settings for the Spotify app\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [POLITICO]\n\nCurrent screenshot is labeled with 29 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 11}}"
}
] |
|
aitw_6915940817252476665_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Check the settings for the Spotify app\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [POLITICO]\nStep 3: Tap [POLITICO]\n\nCurrent screenshot is labeled with 25 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_7030550187439069283_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What time is it in San Francisco?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 13 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 12}}"
}
] |
|
aitw_7030550187439069283_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What time is it in San Francisco?\n\nPrevious actions:\nStep 0: Tap [9]\n\nCurrent screenshot is labeled with 32 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 13}}"
}
] |
|
aitw_7030550187439069283_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What time is it in San Francisco?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [What's a good restaurant in San Diego?]\n\nCurrent screenshot is labeled with 25 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 2}}"
}
] |
|
aitw_7030550187439069283_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What time is it in San Francisco?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [What's a good restaurant in San Diego?]\nStep 2: Tap [X Icon]\n\nCurrent screenshot is labeled with 55 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"What time is it in San Francisco?\"}"
}
] |
|
aitw_7030550187439069283_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What time is it in San Francisco?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [What's a good restaurant in San Diego?]\nStep 2: Tap [X Icon]\nStep 3: Type [What time is it in San Francisco?]\n\nCurrent screenshot is labeled with 25 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"ENTER\"}"
}
] |
|
aitw_7030550187439069283_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What time is it in San Francisco?\n\nPrevious actions:\nStep 0: Tap [9]\nStep 1: Tap [What's a good restaurant in San Diego?]\nStep 2: Tap [X Icon]\nStep 3: Type [What time is it in San Francisco?]\nStep 4: Press [ENTER]\n\nCurrent screenshot is labeled with 32 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_179628004736930330_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 9 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.7, 0.78]}}"
}
] |
|
aitw_179628004736930330_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 23 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 9}}"
}
] |
|
aitw_179628004736930330_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=How+bi]\n\nCurrent screenshot is labeled with 10 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 7}}"
}
] |
|
aitw_179628004736930330_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=How+bi]\nStep 2: Tap [NAV_BAR_RECT Icon]\n\nCurrent screenshot is labeled with 10 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"What's the news in the Philippines?\"}"
}
] |
|
aitw_179628004736930330_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=How+bi]\nStep 2: Tap [NAV_BAR_RECT Icon]\nStep 3: Type [What's the news in the Philippines?]\n\nCurrent screenshot is labeled with 10 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 8}}"
}
] |
|
aitw_179628004736930330_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=How+bi]\nStep 2: Tap [NAV_BAR_RECT Icon]\nStep 3: Type [What's the news in the Philippines?]\nStep 4: Tap [Search or type web address]\n\nCurrent screenshot is labeled with 16 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 11}}"
}
] |
|
aitw_179628004736930330_6 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=How+bi]\nStep 2: Tap [NAV_BAR_RECT Icon]\nStep 3: Type [What's the news in the Philippines?]\nStep 4: Tap [Search or type web address]\nStep 5: Tap [What's the in the Philippines? news]\n\nCurrent screenshot is labeled with 37 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 6) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 27}}"
}
] |
|
aitw_179628004736930330_7 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=How+bi]\nStep 2: Tap [NAV_BAR_RECT Icon]\nStep 3: Type [What's the news in the Philippines?]\nStep 4: Tap [Search or type web address]\nStep 5: Tap [What's the in the Philippines? news]\nStep 6: Tap [| Today's latest from Al Philippines Jazeera]\n\nCurrent screenshot is labeled with 13 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 7) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.57, 0.36]}}"
}
] |
|
aitw_179628004736930330_8 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=How+bi]\nStep 2: Tap [NAV_BAR_RECT Icon]\nStep 3: Type [What's the news in the Philippines?]\nStep 4: Tap [Search or type web address]\nStep 5: Tap [What's the in the Philippines? news]\nStep 6: Tap [| Today's latest from Al Philippines Jazeera]\nStep 7: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 14 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 8) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 10}}"
}
] |
|
aitw_179628004736930330_9 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Philippines?\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=How+bi]\nStep 2: Tap [NAV_BAR_RECT Icon]\nStep 3: Type [What's the news in the Philippines?]\nStep 4: Tap [Search or type web address]\nStep 5: Tap [What's the in the Philippines? news]\nStep 6: Tap [| Today's latest from Al Philippines Jazeera]\nStep 7: Tap [somewhere on the screen]\nStep 8: Tap [We and our partners process data to provide: Use precise geolocation data. Actively scan device characteristics for identification. Store and/or access information device. personalised Create content on a a profile. Select personalised Measure content. content ads, ad measurement, performance. Personalised audience insights and product development.. List of Partners (vendors)]\n\nCurrent screenshot is labeled with 16 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 9) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_65003651068713559_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How big is a lion?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 9 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SWIPE\": \"UP\"}"
}
] |
|
aitw_65003651068713559_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How big is a lion?\n\nPrevious actions:\nStep 0: Swipe [UP]\n\nCurrent screenshot is labeled with 33 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.51, 0.52]}}"
}
] |
|
aitw_65003651068713559_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How big is a lion?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 27 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 9}}"
}
] |
|
aitw_65003651068713559_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How big is a lion?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [google.com/s]\n\nCurrent screenshot is labeled with 25 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"How big is a lion\"}"
}
] |
|
aitw_65003651068713559_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How big is a lion?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [google.com/s]\nStep 3: Type [How big is a lion]\n\nCurrent screenshot is labeled with 15 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"ENTER\"}"
}
] |
|
aitw_65003651068713559_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: How big is a lion?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [google.com/s]\nStep 3: Type [How big is a lion]\nStep 4: Press [ENTER]\n\nCurrent screenshot is labeled with 17 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_16886309439946059005_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Bahamas?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 9 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SWIPE\": \"UP\"}"
}
] |
|
aitw_16886309439946059005_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Bahamas?\n\nPrevious actions:\nStep 0: Swipe [UP]\n\nCurrent screenshot is labeled with 32 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.51, 0.52]}}"
}
] |
|
aitw_16886309439946059005_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Bahamas?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 27 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 10}}"
}
] |
|
aitw_16886309439946059005_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Bahamas?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [google.com/:]\n\nCurrent screenshot is labeled with 24 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"What's the news in the Bahamas?\"}"
}
] |
|
aitw_16886309439946059005_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Bahamas?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [google.com/:]\nStep 3: Type [What's the news in the Bahamas?]\n\nCurrent screenshot is labeled with 15 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"ENTER\"}"
}
] |
|
aitw_16886309439946059005_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Bahamas?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [google.com/:]\nStep 3: Type [What's the news in the Bahamas?]\nStep 4: Press [ENTER]\n\nCurrent screenshot is labeled with 21 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SWIPE\": \"UP\"}"
}
] |
|
aitw_16886309439946059005_6 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Bahamas?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [google.com/:]\nStep 3: Type [What's the news in the Bahamas?]\nStep 4: Press [ENTER]\nStep 5: Swipe [UP]\n\nCurrent screenshot is labeled with 18 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 6) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 13}}"
}
] |
|
aitw_16886309439946059005_7 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the news in the Bahamas?\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [somewhere on the screen]\nStep 2: Tap [google.com/:]\nStep 3: Type [What's the news in the Bahamas?]\nStep 4: Press [ENTER]\nStep 5: Swipe [UP]\nStep 6: Tap [Bahamas - The Independent]\n\nCurrent screenshot is labeled with 15 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 7) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_8166772687237069502_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for hotels in Seattle\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 12 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.65, 0.79]}}"
}
] |
|
aitw_8166772687237069502_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for hotels in Seattle\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 2}}"
}
] |
|
aitw_8166772687237069502_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for hotels in Seattle\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [V_DOWNWARD Icon]\n\nCurrent screenshot is labeled with 31 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"Search for hotels in Seattle\"}"
}
] |
|
aitw_8166772687237069502_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for hotels in Seattle\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [V_DOWNWARD Icon]\nStep 2: Type [Search for hotels in Seattle]\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 8}}"
}
] |
|
aitw_8166772687237069502_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for hotels in Seattle\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [V_DOWNWARD Icon]\nStep 2: Type [Search for hotels in Seattle]\nStep 3: Tap [Search for hotels in Seattle]\n\nCurrent screenshot is labeled with 20 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_12693930241255467158_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Open a new tab in the chrome app\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 11 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"NORMALIZED_COORDINATE\": [0.62, 0.79]}}"
}
] |
|
aitw_12693930241255467158_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Open a new tab in the chrome app\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\n\nCurrent screenshot is labeled with 15 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 6}}"
}
] |
|
aitw_12693930241255467158_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Open a new tab in the chrome app\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=Sea Location]\n\nCurrent screenshot is labeled with 8 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 0}}"
}
] |
|
aitw_12693930241255467158_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Open a new tab in the chrome app\n\nPrevious actions:\nStep 0: Tap [somewhere on the screen]\nStep 1: Tap [google.com/search?q=Sea Location]\nStep 2: Tap [PLUS Icon]\n\nCurrent screenshot is labeled with 21 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_14015533552064702157_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for a new bronzer\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 10 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SWIPE\": \"UP\"}"
}
] |
|
aitw_14015533552064702157_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for a new bronzer\n\nPrevious actions:\nStep 0: Swipe [UP]\n\nCurrent screenshot is labeled with 43 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 5}}"
}
] |
|
aitw_14015533552064702157_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for a new bronzer\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [GOOGLE Icon]\n\nCurrent screenshot is labeled with 46 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 21}}"
}
] |
|
aitw_14015533552064702157_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for a new bronzer\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [GOOGLE Icon]\nStep 2: Tap [nearest sprint store]\n\nCurrent screenshot is labeled with 22 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 1}}"
}
] |
|
aitw_14015533552064702157_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for a new bronzer\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [GOOGLE Icon]\nStep 2: Tap [nearest sprint store]\nStep 3: Tap [X Icon]\n\nCurrent screenshot is labeled with 36 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TYPE\": \"new bronzer\"}"
}
] |
|
aitw_14015533552064702157_5 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for a new bronzer\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [GOOGLE Icon]\nStep 2: Tap [nearest sprint store]\nStep 3: Tap [X Icon]\nStep 4: Type [new bronzer]\n\nCurrent screenshot is labeled with 36 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 5) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 21}}"
}
] |
|
aitw_14015533552064702157_6 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: Search for a new bronzer\n\nPrevious actions:\nStep 0: Swipe [UP]\nStep 1: Tap [GOOGLE Icon]\nStep 2: Tap [nearest sprint store]\nStep 3: Tap [X Icon]\nStep 4: Type [new bronzer]\nStep 5: Tap [sprint nearest store]\n\nCurrent screenshot is labeled with 32 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 6) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SET_TASK_STATUS\": \"COMPLETE\"}"
}
] |
|
aitw_14527044287448665684_0 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the US dollar exchange rateagainst the Swiss Franc?\n\nPrevious actions:\nNone\n\nCurrent screenshot is labeled with 18 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 0) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"PRESS\": \"HOME\"}"
}
] |
|
aitw_14527044287448665684_1 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the US dollar exchange rateagainst the Swiss Franc?\n\nPrevious actions:\nStep 0: Press [HOME]\n\nCurrent screenshot is labeled with 10 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 1) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"SWIPE\": \"UP\"}"
}
] |
|
aitw_14527044287448665684_2 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the US dollar exchange rateagainst the Swiss Franc?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\n\nCurrent screenshot is labeled with 45 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 2) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 5}}"
}
] |
|
aitw_14527044287448665684_3 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the US dollar exchange rateagainst the Swiss Franc?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\nStep 2: Tap [GOOGLE Icon]\n\nCurrent screenshot is labeled with 38 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 3) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 12}}"
}
] |
|
aitw_14527044287448665684_4 | [
{
"from": "user",
"value": "<image>\nImagine you are interacting with an Android device step by step. You can see the screen and know the previous actions through recorded history. The screenshots are labeled with bounding boxes and numeric marks to guide your interactions. Your task is to determine the next action to perform. The possible actions include tapping, swiping, navigating, typing, or setting the task status.\n\nUse the following format for your output, selecting the appropriate key-value pair:\n```json\n{\n \"TAP\": dict, // Use {\"MARK\": mark_index} where `mark_index` is an integer corresponding to the bounding box on the screen, or {\"NORMALIZED_COORDINATE\": [x, y]} where `x` and `y` are float numbers between 0 and 1 representing the normalized coordinates on the screen.\n \"SWIPE\": str, // Choose \"UP\", \"DOWN\", \"LEFT\", or \"RIGHT\"\n \"NAVIGATE\": str, // Choose \"BACK\", \"HOME\", or \"ENTER\"\n \"TYPE\": str, // The text to be typed\n \"SET_TASK_STATUS\": str, // Choose \"COMPLETE\" or \"IMPOSSIBLE\"\n}\n```\n\nCurrent task: What's the US dollar exchange rateagainst the Swiss Franc?\n\nPrevious actions:\nStep 0: Press [HOME]\nStep 1: Swipe [UP]\nStep 2: Tap [GOOGLE Icon]\nStep 3: Tap [good restaurant in miami florida]\n\nCurrent screenshot is labeled with 22 bounding boxes and numeric marks to guide your interactions. Determine your next action (Step 4) and provide the output in the specified dictionary format."
},
{
"from": "assistant",
"value": "{\"TAP\": {\"MARK\": 1}}"
}
] |
End of preview. Expand
in Data Studio
Magma: A Foundation Model for Multimodal AI Agents
Jianwei Yang*1† Reuben Tan1† Qianhui Wu1† Ruijie Zheng2‡ Baolin Peng1‡ Yongyuan Liang2‡
Yu Gu1 Mu Cai3 Seonghyeon Ye4 Joel Jang5 Yuquan Deng5 Lars Liden1 Jianfeng Gao1▽
1 Microsoft Research; 2 University of Maryland; 3 University of Wisconsin-Madison
4 KAIST; 5 University of Washington
* Project lead † First authors ‡ Second authors ▽ Leadership
[arXiv Paper] [Project Page] [Hugging Face Paper] [Github Repo] [Video]
Introduction
This dataset contains the downstream AITW finetuning data used in Magma.
- Downloads last month
- 107