Multi-Agent Control and Intelligent Sensor Allocation With Reinforcement Learning and Genetic Programming

摘要

This project investigated the use of decentralized strategies for control of UAV and UGV swarms. During the first few months we developed an agent-based model of UAVs searching for targets in a pre-determined search area. We tested a variety of control and navigation strategies, including some based on biological principles. Subsequently we develop a second simulator, focusing on the problem of a team of pursuers trying to capture an evader in a 2-D urban environment. The simulator included control strategies for the evader and the pursuers, as well as an interactive world editor for creation of arbitrary urban environments. We used this simulator to run Monte Carlo simulations, obtaining some preliminary statistics on performance of the evader and pursuer strategies.