Find in Library

In this paper presents a new mission planning optimisation method for coverage missions involving Uncrewed Aerial Systems (UAS) and Ground Vehicles (GV) to minimize the mission planning time and the UAS and GV route length. Optimal planning of paths for the UAS and GV using the Mixed Integer Linear Program (MILP), often struggles with computational inefficiency and limited scalability in scenarios with a growing number of waypoints and vehicles. To overcome the MILP computational issues, we present a reinforcement learning technique based on rollout policy optimisation called Multi-Agent Rollout Policy Optimisation (MARPO). Through simulations, we showcase MARPO’s ability to match the precision of conventional MILP formulation in small instances and excel in scalability and computational efficiency in larger cases. Additionally, MARPO is compared with a heuristic method, Multi-Agent Greedy Path-Finding Algorithm (MAGPA), and the superior performance of MARPO in terms of total path length and computational efficiency is demonstrated. Several simulations are presented to showcase the advantages of MARPO. In simulations with 1 UAS and 1 UGV, MARPO achieved path lengths up to 1.56% longer than MILP’s optimum for 9 to 25 waypoints, while significantly reducing computation time by up to 99.88%. In larger scenarios of 36 and 49 waypoints, where MILP was infeasible, MARPO provided convincing solutions with greatly enhanced computational efficiency, demonstrating its robust scalability and effectiveness.