Underwater Glider Path Planning Using Partially Observable Markov Decision Processes

Abstract

Underwater glider has an outstanding advantage of long endurance for ocean observation. This paper presents underwater glider path planning with uncertainties under the framework of partially observable Markov decision processes (POMDPs). The kinematics model and the sensor model of the underwater glider have been built, and the uncertainty of action has been taken into consideration. The priori probability distribution and posteriori probability distribution are obtained from the kinematic model and sensor model, respectively. Particle filtering has been used to combine the two probability distributions. Results show that integrating the uncertainties in state estimation can improve the accuracy of waypoint estimation. Obstacle avoidance is also presented in the same framework.