Multiple Agent Path Planning for Autonomous Area Monitoring

This paper considers informative path planning for two collaborative autonomous agents tasked with monitoring an area against intruders. The area is decomposed into discrete cells that hold the intruder probability distribution and also form the a priori known map over which the agent moves. A computationally efficient intruder model is proposed and the optimal path is formulated in terms of detection probability. A control strategy based on the suboptimal but computationally feasible receding horizon optimization technique is described. The planning is tested in MATLAB on two experimental maps with 108 and 811 cells, respectively. It is concluded that the agents perform rationally and that two agents may almost half the search time compared to a single agent. Real-time performance on ordinary hardware with dedicated software is deemed feasible.