Cooperative-competitive game based approach to the local path planning problem of distributed multi-agent systems

We study the local path planning problem for a distributed cooperative-competitive multi-agent system, where the agents run along a bounded track. All agents are divided into two groups, and each group is required to reach the end point as fast as possible without collision among the agents. During the racing, each agent is interacted with others that it cooperates with its teammates while competes against its opponents. Such interactions among the agents are considered by the game theoretical idea, and then a cooperative-competitive game based model predictive control (CCG-MPC) distributed method is proposed to solve the local path planning problem. Theoretical analysis is provided to show the rationality of the proposed objective function in the path planning problem. Furthermore, we conduct extensive multi-agent racing simulations to compare our CCG-MPC method with the traditional model predictive control (MPC) method, where our CCG-MPC agents show rich behaviors such as avoiding collision with teammates, obstructing and overtaking opponents. Moreover, statistical results show that our CCG-MPC method performs better than the MPC method.