【 title 】A Collaboration of Multi-agent Model using an Interactive Interface

【 The author team 】Jingchen Li, Fan Wu, Haobin Shi, Kao-Shing Hwang

【 Date of publication 】2022.7.13

【 Thesis link 】https://www.sciencedirect.com/science/article/pii/S0020025522007411#!

【 Recommended reasons 】 Multi-Agent Reinforcement Learning Algorithms pay little attention to noisy environments , In this environment , Agents cannot achieve the best strategy training and make correct decisions . This paper studies the influence of noise in multi-agent environment , And propose a multi-agent actor - Critic collaboration (MACC) Model . The model uses lightweight communication to overcome the interference of noise .MACC Each agent in has two strategies ： Collaboration strategy and behavior strategy . The behavior of an agent depends not only on its own state , And it is also affected by other agents through a scalar collaboration value . The collaboration value is generated by the collaboration strategy of each agent , It ensures a concise consensus on the environment . This paper elaborates on the training of cooperative strategies , It also explains in detail how it coordinates behavior strategies in the way of time abstraction mechanism , At the same time, the observation sequence is considered to obtain more accurate perception . Several experiments on the multi-agent collaborative simulation platform show that ,MACC Better performance than baseline in noisy environments , Especially in a partially observable environment .