Safety
How to Train a Cross-Embodiment Robot Navigation Policy with AI Agents
COMPASS is a scalable cross‑embodiment robot navigation framework that builds on NVIDIA’s X‑Mobility pretrained policies, adding lightweight RL fine‑tuning specialists for new robot‑scene pairs to avo
COMPASS is a scalable cross‑embodiment robot navigation framework that builds on NVIDIA’s X‑Mobility pretrained policies, adding lightweight RL fine‑tuning specialists for new robot‑scene pairs to avoid retraining from scratch. An AI agent orchestrates a repeatable workflow—skill validation, asset preparation, training, testing, and approval—across built‑in, SAGE‑10K, and NuRec simulation environments provided by NVIDIA Isaac Sim/Lab, optionally using cuVSLAM for visual SLAM. The resulting policies are packaged as ROS 2 runtime modules that can be deployed on physical robots, enabling rapid, reproducible adaptation to new embodiments and operating conditions.
Related
- GemNav: Discrete-Token Visual Robot Navigation using a Multimodal Large Language Model
- TMRL: Diffusion Timestep-Modulated Pretraining Enables Exploration for Efficient Policy Finetuning
- KITE: Decoupling Kinematics and Interaction for Zero-Shot Cross-Embodiment Manipulation
Source: NVIDIA Developer | 2026-08-26