Design of Autonomous UAV Guidance System Using Monte Carlo Tree Search

In this paper, we propose an online path-planning algorithm using a variation of Monte Carlo tree search (MCTS) for navigating intelligently unmanned aerial vehicles (UAVs) to track mobile ground targets, called the P-UAV algorithm. The proposed algorithm employs a non-myopic method applied to a partially observable Markov decision process (POMDP) model, accounting for long-term decision making. Our algorithm integrates a heuristic technique to efficiently generate paths. The algorithm yields to parallel processing methods to significantly enhance its computational performance, making it suitable for realtime implementation. Simulation experiments show that our path-planning algorithm is efficient and achieves good exploration-exploitation tradeoff in finding a near-optimal solution despite the very large search space.