metadata

pretty_name: ScanBot
license: cc-by-4.0
language:
  - en
tags:
  - robotics
  - instruction-following
  - vision-language
  - laser-scanning
  - multimodal
task_categories:
  - robotics
dataset_info:
  features:
    - name: wrist_camera
      dtype: image
    - name: rgb_gopro
      dtype: image
    - name: instruction
      dtype: string
    - name: tcp_pose
      dtype: string
    - name: joint_states
      dtype: string

ScanBot: Towards Intelligent Surface Scanning in Embodied Robotic Systems

🧠 Dataset Summary

ScanBot is a dataset for instruction-conditioned, high-precision surface scanning with robots. Unlike existing datasets that focus on coarse tasks like grasping or navigation, ScanBot targets industrial laser scanning, where sub-millimeter accuracy and parameter stability are essential. It includes scanning trajectories across 12 objects and 6 task types, each driven by natural language instructions. The dataset provides synchronized RGB, depth, laser profiles, robot poses, and joint states.

📦 Use Cases

Vision-Language Action (VLA)
Instruction-Guided Surface Scanning
3D Surface Reconstruction
Spatial Reasoning and Feature Localization
Laser Profile Analysis for Inspection Tasks

🗂️ Data Description

scanbot/
├── cube1/
│   ├── top_surface/
│   │   ├── path_001/
│   │   │   ├── rgb/
│   │   │   │   └── ...
│   │   │   ├── depth/
│   │   │   │   └── ...
|   |   |   |── 1746226187.997976_gopro.mp4  
│   │   │   ├── robot_joint_states.csv
│   │   │   ├── robot_tcp_poses.csv
│   │   │   └── metadata.json
│   │   ├── path_002/
│   │   │   └── ...
├── cube2
├── cylinder_red
├── cylinder_white
└── ...

💬 Task Levels

Each scan is driven by an instruction that falls under one of the following task categories:

Task Type	Description
T1	Surface Scan
T2	Geometry Focus
T3	Spatial Reference
T4	Functional Target
T5	Defect Inspection
T6	Comparative Analysis

🛠️ Hardware Setup

📖 Version Update

Version 1.0

The initial version of ScanBot contains 197 tasks, and involves 12 different objects.