Unmanned aerial-ground vehicle finite-time docking control via pursuit-evasion games

Cooperation between unmanned autonomous systems has attracted increasing attention in recent years, particularly the challenging problem of unmanned aerial vehicle (UAV) and unmanned ground vehicle (UGV) docking in complex environments with dynamic vehicle interactions. This paper proposes a novel finite-time reinforcement learning control scheme for UAV–UGV docking based on a pursuit-evasion game framework. A pursuit-evasion game formulation is developed where the evader vehicle navigates through complex environments while being pursued by a pursuer vehicle required to track and dock with it. The docking performance is optimized through achieving Nash equilibrium of the pursuit-evasion game. The proposed finite-time reinforcement learning algorithm transforms the value function to finite-time space and employs Actor-Critic neural networks to approximate the value function and optimal controller. A finite-time concurrent learning law is utilized to update the neural network weights, ensuring both the pursuit-evasion game equilibrium and learning process converge within finite time. Lyapunov stability analysis proves the finite-time convergence properties of the algorithm. Experimental validation on an aerial-ground vehicle system demonstrates the effectiveness of the proposed approach in achieving optimal pursuit-evasion performance while maintaining safe landing capability.