Anytime algorithms for multi-agent visibility-based pursuit-evasion games

We investigate algorithms for playing multi-agent visibility-based pursuit-evasion games. A team of pursuers attempts to maintain visibility contact with an evader who actively avoids tracking. We aim for applicability of the algorithms in real-world scenarios; hence, we impose hard constraints on the run-time of the algorithms and we evaluate them in a simulation model based on a real-world urban area. We compare Monte-Carlo tree search (MCTS) and iterative deepening minimax algorithms running on the information-set tree of the imperfect-information game. The experimental results demonstrate that both methods create comparable good strategies for the pursuer, while the later performs better in creating the evader's strategy.