GPU-Accelerated Multi-Objective Optimal Planning in Stochastic Dynamic Environments

The importance of autonomous marine vehicles is increasing in a wide range ocean science and engineering applications. Multi-objective optimization, where trade-offs between multiple conflicting objectives are achieved (such as minimizing expected mission time, energy consumption, environmental harvesting), crucial for planning optimal routes stochastic dynamic environments. We develop multi-objective path planner flows by further developing improving our recently developed end-to-end GPU-accelerated single-objective Markov Decision Process planner. MDPs with scalarized rewards formulated solved idealized environments obstacles. Three simulated scenarios completed to elucidate approach capabilities: (i) an agent moving from start target travel time net-energy consumption when harvesting solar uncertain flow; (ii) and-energy uncertainties obstacle initial positions; (iii) attempting cross shipping channel while avoiding fast ships flow. Optimal operating curves computed fraction the that would be required existing solvers algorithms. Crucially, solution can serve benchmark other approximate AI algorithms such Reinforcement Learning help improve explainability those models.