Maximising Throughput in Lifelong Multi-Agent Pathfinding
Abstract: Many current and emerging industrial applications depend on coordinated teams of agents; to run individual errands, such picking items in a warehouse, and to complete longer and more complex tasks, such as maintenance and manufacturing. At the heart of these systems is a challenging combinatorial problem known as Lifelong Multi-Agent Path Finding. In this talk I will discuss recent progress on this topic from a few distinct perspectives and with an eye toward throughput maximisation; i.e., how to complete the largest number of tasks in the smallest possible time. I will discuss this problem from several perspectives: (i) path planning; (ii) task assignment; (iii) robot dynamics and; (iv) real-time execution. I will highlight a number of recent findings from my own research and I will discuss progress in the broader community by way of the League of Robot Runners; an AI competition initiative that aims to close the gap between abstract planning models and practical application settings.
Bio: Daniel Harabor is an Associate Professor in the Department of Data Science and Artificial Intelligence at Monash University. His research focus is in the area of Heuristic Search with a particular emphasis on the problems and applications of Single- and Multi-Agent Pathfinding. Daniel has more than 80 peer reviewed publications in leading conferences and journals from the area of Artificial Intelligence. He is the founder and co-Chair of the League of Robot Runners Competition, the Grid-based Path Planning^2 Competition and he currently serves as the President of the SoCS Council. In 2025 Daniel was elected as a Senior Member of AAAI.