On-line Learning with Evolutionary Algorithms towards Adaptation of Underwater Vehicle Missions to Dynamic Ocean Environments

Abstract

Autonomous underwater vehicles (AUV) are tasked to ever longer deployments so energy management issues are timely and relevant. Energy shortages can occur due to dynamic ocean conditions that vary temporally and spatially in unpredictable ways. This is compounded by underwater communication challenges. Proposed, is an on-going energy evaluation that assesses the AUV ability to complete the mission through an agent that considers the AUV on-line states, non-linear dynamics, recent learned history, and past history to project an energy shortage. When a shortage occurs an onboard knowledge-based agent re-plans the AUV survey mission using on-line learning with a genetic algorithm given the energy budget, mission duration, and the remaining survey area dimensions. The validated agent is especially effective in the case studied for an energy shortfall resulting from increasing the surveyed area by a factor of 2, for a factor of 2 drop in energy. An agent that effectively monitors and re-plans optimal missions with energy considerations, especially for side scan sonars, is quite novel and increases the operational options of AUVs on long deployments.