Multi-Robot Heterogeneous Adversarial Coverage

Robotic coverage is one of the canonical problems in robotics research, seeking to find a path that visits each point in an area while optimizing some criteria, usually minimizing the time to complete the coverage. This paper considers a variant of the robotic coverage problem, multi-robot adversarial coverage, in which a team of robots is required to cover an area containing threats that might stop the robots with some probability. Motivated by the advantages of using heterogeneous robots for this mission, we formulate the problem while accounting for the trade-off between the coverage time and the expected number of covered cells, considering also the different (heterogeneous) characteristics of the robots involved in the mission. We formulate the problem as a Dec-POMDP and use multi-agent reinforcement algorithms to compute an optimal policy. We have implemented our RL-based methods along with an enhanced heuristic algorithm, and show their superiority compared to the state of the art. Finally, we discuss the possible limitations of learning-based algorithms in different settings.