Adaptive Sampling Based COLREGs-Compliant Obstacle Avoidance for Autonomous Surface Vehicles

Achieving persistent autonomy requires unmanned surface vehicles (USVs) to be able to deal with a wide variety of complex planning situations. In this paper, we introduce a model-predictive, local trajectory planning algorithm for USVs operating in congested and highly dynamic traffic. The planner generalizes the Velocity Obstacle concept to systems with non-linear dynamics, nonholonomic constraints, and any form of low-level feedback controller. High planning performance is achieved by searching in a resolution-adaptive space of vehicle candidate motion goals for a motion goal that optimizes the arrival time to a given global goal. The algorithm gradually identifies, with increasing resolution and focused sampling, an approximate representation of spatiotemporal obstacle regions. The sampling of motion goals can be constrained to ensure the International Regulations for Prevention of Collisions at Sea (COLREGs). We have carried out experiments to demonstrate the practical feasibility of using the planner on a 14-foot (4.3 m) long catamaran USV operating in a harbor-like environment.