What problem does it solve?
Manually onboarding a new robot to COMPASS requires creating multiple specialized configuration files, registering the embodiment in the correct system map, and verifying the robot spawns without simulation errors — a process that is error-prone and requires deep familiarity with the COMPASS codebase structure.
Core Features & Use Cases
- Guided Config Generation: Automatically generates the required ArticulationCfg block and per-embodiment environment config by mirroring existing robot patterns for wheeled, quadruped, or humanoid platforms.
- Automated Registration: Updates the EmbodimentEnvCfgMap in run.py to register the new robot, making it available via the --embodiment CLI flag for training and testing.
- Smoke Test Validation: Runs a controlled test with a single environment to verify the robot spawns at the correct initial pose, does not clip through the floor, and completes the first PPO iteration without joint controller errors.
- Use Case: A robotics researcher wants to add their custom quadruped robot to COMPASS for mobility policy training; this skill walks them through selecting the Spot base pattern, generating the required configs, and running a smoke test to confirm the robot functions correctly in simulation.
Quick Start
Use the compass-newembodiment skill to add your new robot to COMPASS by providing its USD file path, desired lowercase identifier, and matching base robot pattern, then confirm the generated changes and run the smoke test to verify the robot spawns correctly.
