Perceive with confidence: Statistical safety assurances for navigation with learning-based perception

View Full Paper

Ask AI

Bookmark

Authors

Discussion

Analysis shows improved obstacle avoidance in quadruped robots, suggesting robust perception methods enhance reliability.

Obstacle misdetection is significantly reduced by the procedure, enhancing navigation safety against collisions.
In simulation, the approach outperforms uncalibrated models, exhibiting superior success rates in unseen environments.
Using a novel calibration technique, the method ensures robustness against distribution shifts while planning.
Results on a quadruped robot validate empirical safety improvements and reduced conservatism over standard baselines.

Mei et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark