University of Manchester | dda3c: collaborative distributed deep reinforcement learning in swarm agent systems

【 title 】DDA3C: Cooperative Distributed Deep Reinforcement Learning in A Group-Agent System

【 The author team 】Kaiyue Wu, Xiao-Jun Zeng

【 Date of publication 】2022.2.10

【 Thesis link 】https://arxiv.org/pdf/2202.05135.pdf

【 Recommended reasons 】 If multiple agents cooperate to perform their own reinforcement learning tasks , The learning process of each agent can be greatly improved . These tasks may not be exactly the same , But because of the similarity of the task , They still benefit from the communication behavior between agents . in fact , This learning scenario has not been well understood and formulated . As a first job , This paper discusses this scenario in detail , And put forward the group agent Reinforcement learning is the expression of reinforcement learning problems in this scenario , And about single agent And many agent The third kind of reinforcement learning problem of reinforcement learning . This paper proposes that this problem can be solved with the help of modern deep reinforcement learning technology , It also provides a distributed deep reinforcement learning algorithm DDA3C（Discentralised distributed Asynchronous Advantage Actor Critic, Advantages of decentralized distributed asynchrony - Critic ）, It is the first framework designed for group agent reinforcement learning . And in CartPole-v0 Experiments in the game environment show that DDA3C It has achieved ideal performance and good scalability .