Multi-agent Cooperative Pursuit-Evasion Strategies Under Uncertainty

Abstract

We present a method for a collaborative team of pursuing robots to contain and capture a single evading robot. The main challenge is that the pursuers do not know the position of the evader exactly nor do they know the policy of the evader. Instead, the pursuers maintain an estimate of the evader’s position over time from noisy online measurements. We propose a policy by which the pursuers move to maximally reduce the area of space reachable by the evader given the uncertainty in the evader’s position estimate. The policy is distributed in the sense that each pursuer only needs to know the positions of its closest neighbors. The policy guarantees that the evader’s reachable area is non-increasing between measurement updates regardless of the evader’s policy. Furthermore, we show in simulations that the pursuers capture the evader despite the position uncertainty provided that the pursuer’s measurement noise decreases with the distance to the evade.