MORE | Fall 2025

Multi-agent Adversarial Pursuit-evasion Using Deep Reinforcement Learning for UAV Interception

Unauthorized aircraft in controlled airspace pose significant threats to critical infrastructure, requiring autonomous interception systems that can handle evasive targets. While reinforcement learning has shown promise for Unmanned Aerial Vehicle (UAV) control, most approaches train against static or predictable targets that don’t reflect real-world threats. This research develops a multi-agent adversarial framework using Proximal Policy Optimization (PPO) where both pursuer and evader UAVs learn simultaneously. The researcher implemented a custom environment using Stable-Baselines3, beginning with single-agent pursuers before extending to multi-agent scenarios. This approach produces robust interception policies capable of handling intelligent threats, advancing defense systems for airspace security.