No abstract

In many existing multi-agent reinforcement learning tasks, each agent observes all the other agents from its own perspective. In addition, the training process is centralized, namely the critic of each agent can access the policies of all the agents. This scheme has certain limitations since every single agent can only obtain the information of its neighbor agents due to the communication range in practical applications. Therefore, in this paper, a multi-agent distributed deep deterministic policy gradient (MAD3PG) approach is presented with decentralized actors and distributed critics to realize multi-agent distributed tracking. The distinguishing feature of the proposed framework is that we adopted the multi-agent distributed training with decentralized execution, where each critic only takes the agent’s and the neighbor agents’ policies into account. Experiments were conducted in the distributed tracking tasks based on multi-agent particle environments where N(N=3,N=5) agents track a target agent with partial observation. The results showed that the proposed method achieves a higher reward with a shorter training time compared to other methods, including MADDPG, DDPG, PPO, and DQN. The proposed novel method leads to a more efficient and effective multi-agent tracking.

In this paper, the stability problem of a class of multi-agent tracking systems with quasi-cyclic switching topologies is investigated. The existing results of systems with switching topologies are usually achieved based on the assumption that the piecewise constant communication topologies are connected and the switchings are cyclic. The communication topologies are possible to be unconnected and it is difficult to guarantee the topologies switch circularly. The piecewise unconnected topology makes the interactive multi-agent tracking system to be an unstable subsystem over this time interval. In order to relax the assumption constraint, a quasi-cyclic method is proposed, which allows the topologies of multi-agent systems to switch in a less conservative way. Moreover, the stability of the tracking system with the existence of unstable subsystems is analyzed based on switched control theory. It is obtained that the convergence rate is affected by the maximum dwell time of unstable subsystems. Finally, a numerical example is provided to demonstrate the effectiveness of the theoretical results.

No abstract