Master-slave safe cooperative tracking via game and learning based shared control

This work studies the two-player cooperative tracking problem with the players interacting in a master-slave scheme. This problem is formulated as an asymmetric differential game, where the control strategy of the master is unmodifiable and its optimization criterion needs to be recovered. To solve this problem, we develop a learning based algorithm, involving inverse optimization and forward optimal control, to estimate the master cost parameter and design the slave shared controller, simultaneously. A sharing rule based on the estimation of the master cost parameter is proposed, which makes the interaction effective by affecting the control effort paid by the slave directly and continuously. In addition, by using a Lyapunov-like control barrier function, we design a novel safety-critical controller, which can be combined with the shared controller to realize safe trajectory tracking. Some simulation results are given to illustrate the effectiveness of the proposed approaches.