Confluence of Learning and Control Approaches in Multi-Agent Systems

Tuesday July 9, 2024 -- 8:30 am to 4:45 pm -- Pier 9 -- American Control Conference 2024

Abstract

As the world grows increasingly well connected, multi-agent systems have encompassed many critical applications such as cooperative robots, networked control systems, power systems, autonomous vehicles, mobility markets, smart cities, economic institutions, and online social networks. Typically, a multi-agent system comprises many decision-makers that must either learn to act or compute coordinated actions to achieve the design objective. A key feature of such systems is the need for decentralized decision-making arising from different factors such as restricted communication, computational limits, and requirements of resilience against the failure of any subgroup of agents. Under these conditions, traditional centralized approaches for both optimal control and reinforcement learning are rendered unsuitable. Thus, studying the confluence of the different approaches to learning and control in multi-agent systems has emerged as a crucial area of research and development.

This workshop presents recent developments to address the challenges of crafting effective strategies using learning, control, and optimization for multi-agent systems to foster a unified and interdisciplinary understanding of multi-agent systems. The core mission is to stimulate the development of innovative breakthroughs by connecting diverse approaches ranging from, but not limited to, constraint driven control, distributed model-predictive control, ecologically inspired control to team theory and decentralized reinforcement learning. The workshop also explores the intersection of these approaches through the study of complex applications such as multi-human-robot interactions, connected and automated vehicles, persistent monitoring, smart cities, and other cyber-physical systems. As a result, it offers a unique opportunity for researchers and practitioners alike to gain fresh insights, share knowledge and collectively advance the frontiers of multi-agent systems.