MULTI-OBJECTIVE OPTIMIZATION FOR UNMANNED UNDERWATER GLIDERS IN UNCERTAIN ENVIRONMENTS

Abstract

Scientific and strategic interest in unmanned vehicles for persistent ocean monitoring is rising, influenced by concerns over climate and intensifying international politics. One key challenge to their successful implementation is balancing the competing objectives of exploration (information gain) and exploitation (persistent presence). This thesis presents a planning method to balance energy cost and information reward for path optimization for unmanned undersea gliders. Energy planning is accomplished by considering the effects of ocean currents, encouraging vehicle trajectories that follow current vector fields. The vehicle assumes a Gaussian process model representation for the environment, with an updated probability distribution created as the vehicle collects new sensor information. The information reward for a path is quantified as the difference between the prior and posterior distributions as calculated with Kullback-Leibler divergence. By using the ratio of energy cost to information reward as the path optimization metric the planner generates paths that effectively balance energy use with information gain. This method is demonstrated for use with coordinated, multi-vehicle planning. By considering the decremented information reward for previously explored regions, each vehicle plans a separate path through the environment that maximizes its reward based on previously determined paths.