How to intelligently adapt an agent's behaviour to populated, dynamic, and often unpredictable environments is a complex but fundamental question in artificial intelligence.
The world of intelligent systems increasingly consists of autonomous agents that interact in open, dynamic and uncertain environments. In such environments it is often impossible to anticipate on all interactions beforehand. Therefore, intelligent adaptation of behavior among artificial agents is a key research issue in artificial intelligence and agent technology.
Our research concentrates on the multi-agent reinforcement learning (MARL), which receives increasing attention due to the significant advance of reinforcement learning techniques. A reinforcement learning agent learns by interacting with the environment and adapts its strategy to solve sequential decision-making tasks. Under the MARL setups, the research become more challenging as autonomous agents need to learn and adapt to the behavior of other agents as well as the environments.
Our MARL research aims to balance theoretical investigation and practical application. Many problems of our practical interest are concerned with the efficiency and safety of MARL when communication is emerging among agents. For instance, the optimization task when sharing knowledge among different types of agents or when agents interact through a network structure; the exploration task under specific situations such as with safe or human-centric constraints or when the interaction is sparse. Several practical domains are especially interesting for us, including but not limited to automatic driving cars, personalized healthcare system, complex social phenomena and games.