Multi-agent Reinforcement Learning Algorithm

Multi-agent systems are rapidly finding applications in a wide variety of domains such as robotic teams, distributed control, collaborative decision support systems, data mining, etc. 

Although we can program agents to exhibit some basic behaviors, it would be limited with the improvement of task complexity and scene diversity, and many tasks require that agents learn new behaviors online, such that the performance of the agent or of the whole multi-agent system gradually improves. A reinforcement learning (RL) agent learns by interacting with the environment, using a scalar reward signal as performance feedback. The simplicity and generality of this setting make it attractive also for multi-agent learning. We are committed to studying some challenges of Multi-Agent Reinforcement Learning (MARL), and solving application problems based on MARL.