Orchestrate COMPASS robot navigation training from SAGE-10k scene search to OSMO cloud submission.
npx skills add https://github.com/NVlabs/COMPASS --skill compass-nvlabs
Please help me install this Agent Skill. Skill: compass Source: https://github.com/NVlabs/COMPASS/tree/main/.claude/skills/compass Command: npx skills add https://github.com/NVlabs/COMPASS --skill compass-nvlabs
This Skill eliminates the fragmented, manual work of setting up and running COMPASS cross-embodiment mobility policy training, unifying scene search, USD conversion, scene registration, training, evaluation, and cloud submission into a single guided workflow to reduce setup time and avoid costly errors from broken scenes or misconfigured commands.
Use the compass skill to train a cross-embodiment navigation policy for a bedroom scene by searching SAGE-10k for a matching layout, converting it to USD, verifying the scene in Isaac Sim, and launching a full training run with a preview check for confirmation.
High-intent search queries and answers about installing and using this skill.
Cross-embodiment robot navigation policy training with SAGE-10k involves searching the dataset by natural language, converting matching scenes to USD format, and registering them for residual RL training. The compass skill orchestrates this end-to-end workflow including evaluation and cloud submission.
Yes, SAGE-10k dataset scene search supports natural language descriptions to find matching layouts for robot navigation training. The compass skill handles searching 10,000 scenes, converting results to USD format, and registering them for COMPASS training workflows without manual scene generation.
Isaac Sim visual verification is required as a mandatory gate to catch invalid scenes or robot spawn issues before launching expensive full-scale residual RL training. The compass skill enforces this verification step to prevent GPU-intensive training on broken scenes.
Submitting robot navigation training workloads to OSMO cloud clusters is streamlined through the compass skill, which handles container and workflow configuration automatically. This enables scaled training directly from the guided pipeline without manual cloud setup or configuration steps.
USD scene conversion transforms SAGE-10k matching scenes into Isaac Sim compatible format, followed by registration for COMPASS training. The compass skill automates this pipeline execution, eliminating manual context switching between scene management and training configuration steps.
A preview training run provides mandatory visual verification to catch invalid scenes or robot spawn issues before launching expensive full-scale training. The compass skill enforces this gate to prevent GPU-intensive training on broken scenes and requires user confirmation before proceeding.
hkqaiGenerate Isaac Lab simulation scripts from YAML scene descriptions.
robotics-playgroundGuides RL locomotion policy development for Unitree G1 using NVIDIA Isaac Sim and Isaac Lab.
mashharukiRun Python workloads on managed Hugging Face Jobs infrastructure.
Kling0012Create and train AI learning plugins with AgentDB reinforcement algorithms.
