Safe Robot Navigation

Safe robot navigation studies how robots can reach goals without collisions while accounting for limited sensing, uncertain motion, changing obstacles, and hazards such as people or difficult terrain. Common approaches include control barrier functions and reachability analysis for enforcing safety constraints, model predictive control and graph search for planning, and learned visual or language-guided policies for interpreting unfamiliar scenes. Research increasingly combines these methods, using semantic risk maps, human-motion prediction, uncertainty calibration, or runtime safety filters to make decisions in dynamic and partially observed settings. A central challenge is balancing formal guarantees and reliable evaluation with the adaptability and efficiency of learned methods, which can become overly cautious or fail when predictions, timing, or real-world conditions differ from assumptions.