Heterogenous vehicle routing: comparing parameter tuning using genetic algorithm and bayesian optimization

Abstract

Heterogeneous vehicles (e.g., unmanned ground vehicle (UGV) and unmanned aerial vehicle (UAV)) are best suited for surveillance application over large areas. UAVs are fast, but fuel limited, while, UGVs have a larger fuel capacity, but are relatively slow. When UAVs are combined with UGVs they can provide larger coverage at a relatively fast speed. The UAV may also be recharged on the UGV as needed. The resulting route optimization problem is computationally complex, but may be solved relatively fast using heuristics. In this paper, we solve for a mission route using a two-level optimization; (1) the UGV route is assigned using heuristics with free parameters, (2) the UAV route is solved using a vehicle routing problem formulation with capacity constraints, time windows, and dropped visits. However, this open-loop two-level optimization may yield non-optimal solutions or fail completely because of poor choice of UGV parameters. Our primary objective is to explore closed loop optimization where the free parameters of the UGV routes are optimized using Bayesian optimization and Genetic algorithms. Our results show that both methods produce good quality solutions, but bayesian optimization is computationally more efficient than genetic algorithm.