Real-time collision avoidance for robots in tight, crowded spaces

Autonomous robots need to avoid collisions instantly using only their own sensors, especially in constrained spaces like underground tunnels. This work combines 3D LIDAR perception with control barrier functions—mathematical constraints that prevent collisions—applied directly at the control frequency. The safety region adapts to the robot's orientation as a time-varying ellipsoid, handling dynamic obstacles from point cloud data without disrupting the robot's primary task. Field experiments on a quadruped platform demonstrate reliable operation through narrow corridors, around moving obstacles, and under localization errors.