Towards a COLREGS compliant autonomous surface vessel

Abstract

Bringing about Autonomous Surface Vessel (ASV) autonomy faces two major challenges in the standard robot navigation pipeline - mapping the world and making decisions based upon that map; for the most part, ASV state estimation and vehicle control may be assumed due to the reliable presence of GPS on the ocean and the scale on which decisions are currently made. Numerous approaches have been taken at solving these problems, yet few full systems have been demonstrated in complex live harbor environments. In this paper we describe a full system for navigating these environments using the WHOI Jetyak. The Jetyak is equipped with a broadband marine radar and Inertial Navigation System (INS) in order to perform static and dynamic obstacle detection. Once detected, individual measurements are combined over time using data association to identify and track specific vehicles. Most radar manufacturers come with this capability through automatic radar plotting aids (ARPA), however, these are typically not exposed to the user for work in robotics. We use a common pipeline of data association between scans using a Joint Probabilistic Data Association (JPDA) algorithm to give unique ids to boats, and an Unscented Kalman Filter (UKF) to track and predict their motion. With a projection of where other vehicles in the environment are and a prediction of where they are going, the ASV must make decisions on what actions to take to avoid other boats. While this problem is similar to the common 2D+Time obstacle avoidance problem often seen in terrestrial robotics, it has the added element that all boats must act under the constraints of COLREGs. Similar to the rules of driving, COLREGs defines which boat has the right-of-way, and what actions the non-right-of-way boat must perform. We account for these rules using a reactive path planning algorithm based on the one proposed by Kuwata et. al., that incorporates the rule based constraints of COLREGS into a velocity obstacles algorithm. In addition we test an implementation of incorporating COLREGS into a global motion planner to provide better performance in complex environments.