Pixels to Pursuit: Reinforcement Learning-Based Control Strategies for UAV-AGV Collaborative Systems

This paper presents reinforcement learning-based control strategies for vision-based UAV tracking by AGVs. We developed custom YOLOv8 object detection models achieving 98 % accuracy and implemented DQN and DDPG controllers trained in simulation environments. Experimental validation demonstrates DDPG outperforms PID, SMC, and DQN controllers with superior tracking accuracy and reduced settling time. The approach enables robust UAV-AGV collaboration in GPS-denied environments with potential applications in agriculture and search-and-rescue operations.