Abstract

Multi-agent systems are a core area of AI where multiple autonomous agents coordinate and collaborate to solve complex problems. This talk will present our research across two key approaches to multi-agent coordination. First, we will discuss our work in multi-agent reinforcement learning (MARL), including state-of-the-art algorithms for complex coordination challenges and their applications in domains such as multi-robot warehouses and strategic games like Starcraft and Go. Second, we will explore recent advances in LLM -based multi-agent systems, where large language models enable dynamic multi-agent teams to collaborate on complex workflows and tackle real-world problems. We will share our vision for a Manager Agent designed to achieve scalable and compliant orchestration of human-AI teamwork.