Related papers for multi-agent reforcement learning.
-
Multi-Agent Coordination in Adversarial Environments through Signal Mediated Strategies
设计了two-player中心化采样缓冲池(该buffer可以利用彼此的动作进行训练),设计了信号中介策略框架(每个agent的策略在每个episode开始前condition on一个信号,该信号表征多智能体策略之间的均衡)
-
Signal Instructed Coordination in Cooperative Multi-agent Reinforcement Learning
受到相关均衡CE的启发,设计一个中心信号控制器,每个agent的策略都依托于这个隐含全局信息的信号中;主要学习目标有三个:对齐智能体策略和协作信号、减轻其他智能体策略的不确定性以降低协作难度、学到多样性策略;
-
Attention Actor-Critic algorithm for Multi-Agent Constrained Co-operative Reinforcement Learning
受限MARL一方面要最小化累积cost,另一方面要满足约束;文中提出训练一个对偶拉格朗日的方式,以拉格朗日乘子权重作为约束的权重去求解最优策略;
-
Modeling the Interaction between Agents in Cooperative Multi-Agent Reinforcement Learning