Pixels to Pursuit: A Comparative Study of PID, SMC, DQN, and DDPG based control for Vision-Based UAV-AGV Collaboration(Under Review )

Hetrogeneous multiagent collaborative systems have gained significant attention for it’s potential to enhance efficiency and autonomy in various safety surveillance and rescue operations. Tracking motion between Unmanned Aerial Vehicle (UAV) and Autonomous Ground Vehicle (AGV), landing of UAV over AGV are fundamental collaborative tasks. Tracking a UAV using an AGV presents greater challenges compared to tracking an AGV with a UAV. This is primarily due to the AGV’s limited spatial maneuverability. Vision based tracking of UAV by an AGV is considered in this work to do a comparative study of three distinct control strategies Proportional-IntegralDerivative (PID) controllers, Sliding Mode Controllers (SMC), and Reinforcement Learning (RL)-based controllers—applied to a UAV-AGV multi-agent collaborative system. The primary objective is to evaluate the performance, robustness, and adaptability of these controllers in managing the complex dynamics and interactions between UAVs and AGVs during collaborative tasks. Our proposed methodology incorporates deep learning-based UAV detection. Autonomous vision-based control for AGV is developed, guided by the extracted image features. We present experimental validations for PID, SMC and RL based control of the AGV in outdoor conditions. The experimental results are discussed for performance analysis of the approach. This comparative analysis provides valuable insights into the selection of appropriate control strategies for UAV-AGV systems, contributing to the advancement of autonomous multi-agent operations in various applications.