Multi-agent systems comprise multiple decision-making agents which act and interact autonomously in a shared environment towards achieving specified tasks, such as ISR missions. Each agent is a software system with interfaces to sense potentially noisy and incomplete information from the environment, and to choose actions within the environment such as high-level planning decisions (resource allocation and coalition formation) or low-level controls (movements of mobile assets).

This theme seeks to develop new algorithms for agent decision-making in complex, dynamic multi-agent systems to enable efficient and scalable coordination of agents.

A strong focus will be on data-driven methods based on deep reinforcement learning (RL) and multi-agent RL, whereby agents learn coordinated decision policies by autonomously exploring a complex action space and identifying optimal coordination strategies.

Other important method components will include dynamic team formation in “ad hoc” scenarios where agents must collaborate with new, previously unknown agents on-the-fly.

Co-I: Prof Ram Ramamoorthy