Path planning and collision avoidance for safe autonomous vessel navigation in dynamic environments

This work presents a grid based path planning algorithm to generate evasive trajectories for vessels in situations that potentially lead to a collision. For an operation in the presence of civil marine traffic, the evasive behaviour has to obey local and international laws. Therefore, a specialised A* search algorithm is used to find a sequence of collision-free and reachable waypoints taking into account the International Regulations for Preventing Collisions at Sea, 1972 (COLREG), information from Electronic Nautical Charts (ENCs) and other vessels detected by a Radar based obstacle detection systems. To guarantee the reachability of these estimated waypoints, the kinematic constraints of the own vessel has to be considered by the A* algorithm. Therefore, two new cell neighbourhoods for grid based search algorithms, the so called region of reachability and the T-Neighbourhood, are introduced. Furthermore, a trajectory for an autonomous driving vessel should be smooth with a continuous curvature to improve the motion control performance. Thus, the resulting sequence of waypoints is interpolated using Bézier curves. Ship domains are used to define a safety distance to other vessels. When applying such a collision avoidance system for real ships, it is, however, to note that both the sensors and the models are imperfect. To take this into account, the grid based collision avoidance is extended to consider the measuring and tracking uncertainties of the obstacle detection system. Several tests are performed on the Lake Constance with an automated recreational craft to estimate typical navigation parameters such as a proper distances other vessels. The results of these tests are used to optimise the A* search parameters further.