Uncertainty aware path planning and collision avoidance for marine vehicles

Motion planning, a challenging problem for any autonomous agent, becomes even more difficult for marine craft due to under-actuation, nonlinear and unmodelled kinematics, uncertainties and noise in sensor data, uncertain obstacles, wind and waves. We consider a marine craft with unmodelled kinematics, subject to environmental disturbances and in the
presence of moving obstacles with unknown kinematics. We utilise a Luenberger observer structure to estimate the marine craft and obstacles kinematics in real-time using sensor data. We furthermore bound the estimation error and subsequently use it explicitly in the determination of the guidance control laws. The modular nature of this algorithm enables integration with existing state-of-the-art path planning methods. The effectiveness of our proposed approach is illustrated and compared using Imazu benchmark scenarios and several
existing planning methods, specifically, velocity obstacle method, geometric line-of-sight (LOS), time-critical LOS based guidance methods (finite-time), assuming unmodelled kinematics of the marine and obstacles craft, while being exposed to wind effects.