Robot Docs
Robotics simulation and perception stack for the Unitree G1 humanoid robot. Autonomous navigation with Nav2, GPU-accelerated perception with Isaac ROS, and reinforcement learning locomotion trained in Isaac Lab.
Carter robot warehouse demo with Nav2 and Foxglove Studio
Unitree G1 locomotion policy trained with RL in Isaac Lab
Unitree G1 in the ISRMM lab during a motherboard swap
Stack
| Component | Technology |
|---|---|
| Simulator | NVIDIA Isaac Sim 5.1 |
| Perception | Isaac ROS 4.1 (cuVSLAM, AprilTag, nvblox) |
| Navigation | Nav2 with MPPI controller |
| RL Training | Isaac Lab (RSL-RL) |
| Robot | Unitree G1 humanoid |
| Workstation | Linux, RTX 3090 |
| Visualization | Foxglove Studio |
Features
- Autonomous navigation — Nav2 with SMAC Hybrid-A* planner and MPPI controller in a warehouse environment
- GPU perception — cuVSLAM visual odometry and AprilTag detection running in Isaac ROS Docker
- RL locomotion — Unitree G1 walking policy trained from scratch with reinforcement learning in Isaac Lab
- Remote streaming — Foxglove Studio over Tailscale for real-time sensor visualization
- Full documentation — architecture diagrams, setup guides, and tuning notes for every component