Multi-UAV-UGV Collision-Free Tracking Control via Control Barrier Function-Based Reinforcement Learning

This paper introduces a novel hierarchical control approach for feature matching, real-time tracking and inter-UAV collision avoidance in multiple unmanned aerial vehicle-unmanned ground vehicle (multi-UAV-UGV) collaborative tracking. Our approach divides into three layers: optimal feature matching, tracking control by reinforcement learning (RL), and collision avoidance using control barrier functions (CBFs). First, a distance cost matrix is cleverly constructed based on the feature matching capabilities of UAVs and UGVs to determine the optimal matching configuration. It allows UAVs to perform the tracking task while minimizing travel distance. Second, a RL-based tracker is developed to achieve precise real-time tracking without depending on UAV dynamic models. The tracker is trained in a single UAV-UGV environment, which reduces policy convergence difficulty by simplifying state space and interactions compared with training in complex multi-UAV-UGV scenarios. Third, a collision avoidance mechanism based on CBFs is introduced to transform RL commands into collision-free actions by solving a quadratic programming (QP) problem. Extensive simulations and real-world experiments demonstrate the effectiveness of the proposed approach.