Abstract: Games and puzzles (one-player games) play an important role in many research areas, including computer science. Chess in particular received considerable attention as one of the original "grand challenge" problems of AI research. The 75 years of board and card game research shows an interesting evolution of the ideas and technologies used to achieve high -- even super-human -- performance in these domains. The speaker has been involved in AI technology applied to games for over four decades, and gives his perspective on the past, present, and some thoughts on the future of games in AI research.

Bio: Jonathan Schaeffer is a professor in the Department of Computing Science. His research is in the area of artificial intelligence, with much of his work being demonstrated using games and puzzles. His checkers-playing program Chinook was the first computer to win a human world championship (1994), a feat recognized in the Guinness Book of World Records. He has helped develop superhuman programs at poker (two-player limit Texas Hold'em -- also in the Guinness Book of World Records). He is a Fellow of the Association for the Advancement of Artificial Intelligence (AAAI) and a Fellow of the Royal Society of Canada. He is co-founder of Onlea (onlea.org) that creates engaging online learning experiences.

Abstract: Multi-Agent Path Finding (MAPF) is the problem of finding paths for a set of agents such that each agent reaches its goal while avoiding collisions with static obstacles and the other agents. MAPF has drawn significant attention in the passing decade, and significant progress has been made in solving it, including significant contributions from the Combinatorial Search research community. However, many of these algorithmic contributions have been focused on the classical version of the problem, in which agents are assumed to operate in a discrete, deterministic, and fully observable environment. The question arises: which ideas from classical MAPF can transfer to more complex environment where these assumptions do not hold, and when new ideas are needed. In this talk, I will survey some of the progress made by myself and others on answering this question. This includes algorithms for solving MAPF when time is not discretized such as Continuous Conflict-Based Search (CCBS), offline and online approaches for handling actions with non-deterministic or stochastic effects, and algorithms for handling uncertainty over the underlying environment. Finally, I’ll describe a recently proposed framework for using MAPF algorithms in a realistic environment where agents must operate in real-time and be robust to planning failures.

Bio: Roni Stern is an Associate Professor of Software and Information Systems Engineering in Ben Gurion University of the Negev. He heads the Software Engineering program and is the co-PI of the Search, Planning, and Learning (SPL@BGU) lab and the Anomaly Detection and Diagnosis (AiDnD) lab. In addition, he serves as an Associate Editor in JAIR and in charge of the “conference award-winning papers” track. In the past, he served as the president of the Symposium on Combinatorial Search (SoCS), a Principal Scientist at the Palo Alto Research Center (PARC), and held various software engineering roles. He earned his Ph.D. from Ben Gurion University, under the supervision of Prof. Ariel Felner and Prof. Meir Kalech.