Multi-agent pursuit with decision-making and formation control

This article is concerned with pursuit of multiple evaders with a group of mobile robots. We address this problem from a novel stochastic point of view. Objective of the pursuit is to track and eventually capture the evaders. The evaders possess certain level of intelligence in the way of changing their motion pattern. This necessitates group decision making and coordination of the pursuers. A decision-making scheme of formation changing for the pursuers is proposed in this paper. The pursuers track the targets using a leader-following strategy with the state of the evaders being estimated with distributed Kalman filtering. When the motion pattern of the evaders varies, the pursuers change their formation accordingly based on the Decentralized Partially Observable Markov Decision Processes (DEC-POMDP). Decisions are made according to the pursuers´ confidence of their estimation on the evaders and seek to approach maximal reward. Simulation results show the feasibility and validity of the proposed scheme.