X-NavDP: Generalizing Navigation Diffusion Policy to Novel Behavior and Embodiments with Group Q-score Reweighted Matching

Researchers introduced X-NavDP, a diffusion-policy RL method for visual navigation across different robot bodies and difficult scenarios.

Its GQRM post-training framework combines behavior-perturbed exploration with group Q-score normalization for reweighted score matching.

With distributed online RL, the policy improved cross-embodiment navigation in both simulation and real-world hard cases.

The work addresses a major generalization gap in navigation diffusion policies and points to more adaptable robot navigation.