Thinker-IL-LAB

Thinker is an imitation learning toolchain for humanoid robots. It provides an end-to-end pipeline from simulation data collection to real-robot deployment for the Walker TienKung Pro, Walker TienKung Dex, Walker S2 EDU, and Walker C1 EDU platforms.

Built on NVIDIA Isaac Sim and HuggingFace LeRobot.

GitHub Repository · LQX-Code-SH/Thinker-IL-LAB Get Started

Capabilities

Capability	Description	Status
ROS simulation	High-fidelity Isaac Sim environment for teleoperation and data collection	Available
Teleoperation	Keyboard / space-mouse teleoperation with a unified sim-to-real interface	In progress
Data collection & conversion	HDF5 / LeRobot data collection, conversion, and cleaning	Available
Policy training	Imitation learning with LeRobot (ACT / Pi0.5)	Available
Real-robot deployment	Policy inference and robot control, including a warm-started inference server	Available
Supported robots	Walker TienKung Pro / Walker TienKung Dex / Walker S2 EDU / Walker C1 EDU	Released

Gallery

Simulation preview

Walker TienKung Pro

Walker TienKung Dex

Walker S2 EDU

Walker C1 EDU

Deployment results

Walker TienKung Pro · Simulation

Walker TienKung Pro · Real robot

Walker S2 EDU · Simulation

Walker S2 EDU · Real robot

Walker TienKung Dex · Simulation

Walker TienKung Dex · Real robot

Walker C1 EDU · Simulation

Walker C1 EDU · Real robot

Application case

Walker S2 EDU · VLA pick-and-place

See Walker S2 EDU application case.

Architecture

The repository contains two sub-projects: ubt_sim (simulation platform) and ubt_IL (imitation learning platform).

┌──────────────────────────────────────────────────────────────────────────┐
│                          ubt_IL imitation learning                       │
│              data conversion ──► training ──► evaluation ──► deployment   │
└───────────▲───────────────┬──────────────────────▲──────────────┬────────┘
            │               │④ sim deployment       │              │⑤ real deployment
            │① data          ▼ (ROS bridge)          │② real data    │ (ROS bridge)
   ┌────────┴────────────────┐        ┌─────────────┴──────────────▼──────┐
   │      ubt_sim            │        │ Real robots:                      │
   │   (Isaac Sim container) │        │ Walker TienKung Pro /             │
   │                         │        │ Walker TienKung Dex /             │
   │                         │        │ Walker S2 EDU / Walker C1 EDU     │
   │                         │        │ ROS_DOMAIN_ID=0 direct network    │
   └─────────────────────────┘        └───────────────────────────────────┘

See Architecture Overview for details.

Start here

• Getting Started - Clone the repository, check requirements, and follow the end-to-end flow
• Simulation Platform - Isaac Sim, teleoperation, and data collection
• Imitation Learning Platform - Data conversion, training (ACT / Pi0.5), and evaluation
• Walker TienKung Pro - Simulation / conversion / training / deployment / ARM board deployment
• Walker TienKung Dex - Single-arm dexterous manipulation: simulation / training / real-robot deployment
• Walker S2 EDU - Simulation / conversion / training / deployment / inference server
• Walker C1 EDU - Dual-arm dexterous manipulation: simulation / training / real-robot deployment

Acknowledgements

This project builds on the following open-source projects:

• NVIDIA Isaac Sim - High-fidelity robotics simulation
• HuggingFace LeRobot - Imitation learning framework (ACT, data format, training and inference tools)
• Thinker Studio - Thinker teleoperation data collection platform
• Thinker Cosmos Open Source Community - Thinker open-source community

License

Apache-2.0