Negotiating Visibility for Safe Autonomous Navigation in Occluding and Uncertain Environments

Abstract

Navigation through an occluded environment is a challenging task for autonomous mobile robots (AMR), since they must balance both safety and speed in an attempt to fluidly steer around occlusions in uncertain environments. This is because real world environments have dynamic actors that may be occluded to the robot during motion, introducing uncertainty. One key element of eliminating this uncertainty is moving in such a way to maximize perception around these occlusions. This letter presents a novel control framework that combines both perception and safety constraints, resulting in motion that is quick and safe when occlusions are present. Perception is satisfied using a model predictive control (MPC)-based approach to provide inputs that increase visibility around occlusions while safety is promoted by modeling uncertainties as projected probabilities of occupancy derived from current observation and expected traffic motion. Improvements in visibility, safety, and speed are shown in simulations and are experimentally validated using an unmanned ground vehicle.