Trajectory planning with adaptive control primitives for autonomous surface vehicles operating in congested civilian traffic

We introduce a model-predictive trajectory planning algorithm for unmanned surface vehicles (USVs) operating in congested civilian traffic. The planner reasons about the availability of contingency maneuvers needed in case of any of the civilian vessels breaches the International Regulations for the Prevention of Collisions at Sea (COLREGs). Our exploratory study indicated that implementing the envisioned planner requires significant speed up of trajectory planning to cope with the dynamics of the scene, and evaluation of collision risk. We describe a new method for efficiently searching 5D state space for a dynamically feasible trajectory using adaptive control action primitives. The algorithm estimates the congestion of the state space regions to evaluate collision risk, and then dynamically scales action primitives used during the search while preserving their dynamical feasibility. Our simulation experiments demonstrate that this leads to a substantial increase in the search efficiency and a decrease in the number of collisions, especially in complex scenarios with a higher number of civilian vessels.