How fan-ziqi/robot_lab works
Project type and purpose
• Project type: A robotics‑focused RL extension library (not a web app)
• Purpose: Provide Gym‑style learning environments for various robots (quadruped, humanoid, wheeled) that run inside IsaacSim (Omniverse) via IsaacLab extension.
• Developer Shell → invokes CLI scripts
• CLI scripts import robot_lab extension & third‑party libraries
• robot_lab code registers environments with Gym, instantiates ManagerBasedRLEnv
• ManagerBasedRLEnv ↔ exchanges state/action data with IsaacSim
• RL Runner (rsl_rl_ppo or AMP variants) ↔ ManagerBasedRLEnv
• Simulation reads USD assets / URDF conversion scripts produce USD assets
• RL Runner → writes logs and checkpoints
• TensorBoard reads logs
• Docker Compose orchestrates containers: simulation + services
Architectural patterns & design principles
• Plugin‑based extension: robot_lab loaded by IsaacLab’s extension manager
• Modular environment registration: one folder per robot/config, Gym registry
• Separation of concerns:
– CLI layer (scripts/)
– Core library (source/robot_lab)
– Third‑party algorithms (amp_utils, rsl_rl_amp)
– Data assets and config files
• Configuration via code and TOML/YAML (configclass decorator)
Detail level
• Don’t clutter with every single file—focus on logical groupings (scripts, core, assets, third‑party, deployment)