SaiResearch commited on
Commit
10f4622
·
verified ·
1 Parent(s): 9d33299

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +81 -1
README.md CHANGED
@@ -10,4 +10,84 @@ tags:
10
  - reinforcement_learning
11
  size_categories:
12
  - n<1K
13
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
  - reinforcement_learning
11
  size_categories:
12
  - n<1K
13
+ ---
14
+ # Booster T1 Dataset
15
+
16
+ The **Booster T1 Dataset** is a collection of motion and control data capturing a humanoid robot (Booster T1) performing a diverse set of soccer-related actions. These include skills necessary for robot soccer such as kicking, dribbling, and goal kicks.
17
+
18
+ This dataset is designed to support research in **robot soccer, reinforcement learning, motion planning, imitation learning, and control of bipedal robots** in dynamic, contact-rich environments.
19
+
20
+ ---
21
+
22
+ ## Dataset Details
23
+
24
+ ### Dataset Description
25
+
26
+ - **Curated by:** ArenaX Labs
27
+ - **License:** MIT
28
+ - **Format:** `.npz` files containing robot kinematic and dynamic states
29
+ - **Purpose:** Provide expert demonstrations and trajectories for training and benchmarking soccer-playing humanoid robots.
30
+
31
+ ---
32
+
33
+ ## Uses
34
+
35
+ ### Direct Use
36
+
37
+ - Training reinforcement learning and imitation learning policies.
38
+ - Motion planning and control benchmarking for humanoid soccer.
39
+ - Studying dynamic skills like ball-kicking, goal-kicking, and repositioning.
40
+ - Curriculum learning: starting from balance and stepping, progressing to soccer maneuvers.
41
+
42
+ ### Out-of-Scope Use
43
+
44
+ - Human motion modeling or biomechanical studies (data is robot-specific).
45
+ - Applications outside robotics locomotion and soccer (e.g., medical or sensitive domains).
46
+ - Any use that attempts to infer personal, demographic, or identity-related data (not present in this dataset).
47
+
48
+ ---
49
+
50
+ ## Dataset Structure
51
+
52
+ Each `.npz` file contains the following arrays:
53
+
54
+ - **qpos**: Concatenated positions (root position, root orientation quaternion, and DOF positions).
55
+ - **qvel**: Concatenated velocities (linear velocity, angular velocity, and DOF velocities).
56
+ - **xpos, xquat, cvel, subtree_com, site_xpos, site_xmat**: Currently placeholder arrays (`zeros`) reserved for extended features such as body/site positions and COM.
57
+ - **split_points**: Start and end indices for trajectory segmentation.
58
+ - **joint_names**: Names of robot joints.
59
+ - **frequency**: Target control frequency of the recorded trajectory.
60
+ - **njnt**: Number of joints.
61
+ - **jnt_type**: Joint types (0 = root, 3 = hinge).
62
+ - **body_names, site_names, metadata**: Reserved metadata placeholders.
63
+ - **body_*** and **site_*** arrays: Empty placeholders for MuJoCo-style body/site information (position, orientation, weld IDs, etc.).
64
+
65
+ This structure allows loading trajectories directly into MuJoCo-compatible formats for playback or analysis.
66
+
67
+ ---
68
+
69
+ ## Dataset Creation
70
+
71
+ ### Curation Rationale
72
+
73
+ The dataset was created to provide a standardized benchmark of soccer-related skills for humanoid robots, facilitating progress in robotic soccer research.
74
+
75
+ ### Recommendations
76
+ - Use as a **benchmark for policy learning** rather than as a standalone dataset for generalization.
77
+ - Combine with simulated data augmentation for robustness.
78
+
79
+ ---
80
+
81
+ ## Citation
82
+
83
+ If you use this dataset, please cite as:
84
+
85
+ **BibTeX**
86
+ ```bibtex
87
+ @dataset{arenax2025booster,
88
+ title = {Booster T1 Dataset},
89
+ author = {ArenaX Labs},
90
+ year = {2025},
91
+ publisher = {Hugging Face},
92
+ url = {https://huggingface.co/datasets/SaiResearch/booster_dataset}
93
+ }