MSE 298 Seminar: Task Allocation in Teams of Autonomous Ground Vehicles and Humans

McDonnell Douglas Engineering Auditorium (MDEA)
Bogdan I. Epureanu, Ph.D.

Department of Mechanical Engineering
Department of Electrical Engineering and Computer Science
Director, Automotive Research Center
University of Michigan, Ann Arbor

Abstract: Autonomous vehicles are increasingly thought of as team members alongside humans in both military and civilian applications. Such autonomous agents are capable of handling dangerous tasks but are limited in their reactions to unforeseen events. At the same time, humans have more adaptive and creative problem-solving skills but are limited in terms of handling some specific tasks and managing cognitive loads. The inclusion of autonomy within a team requires a significant effort to train agents to distribute tasks dynamically for optimal operation. In addition, the need to address diverse requirements requires heterogeneous teams where agents can have different capabilities and levels of risk tolerance. While heterogeneity in multi-agent teams offers benefits, new challenges arise including how to find optimal heterogeneous team compositions and how to distribute tasks dynamically among agents in complex operations.

This presentation will discuss modeling tools for autonomous vehicles with focus on artificial intelligence algorithms for heterogeneous agents to learn dynamically task distributions in teams with humans and other autonomous agents through reinforcement learning in synthetic environments. The approach extends Decentralized Partially Observable Markov Decision Processes to be compatible to model various types of heterogeneity. A disaster relief scenario simulated in a high-fidelity game engine environment is discussed where a human interacts with the environment in real-time using virtual reality. An adaptive algorithm is created to assist the humans in their decision-making and improve their performance by continuously evaluating the human’s cognitive task loads. The heterogeneity of the team is described by the differences in agent capabilities of task handling, sensing and communication, as well as the level of risk aversion in humans’ decision-making processes. Results of this study show that the trained autonomous agents using the developed algorithms can reliably collaborate with humans.

This work is part of the activity of the Automotive Research Center (ARC), which is the U.S. Army Center of Excellence in the area of modeling and simulation (M&S) of ground vehicles led by the University of Michigan. The ARC focuses on M&S of heterogeneous multi-vehicle teams of humans and autonomous vehicles capable to adapt in adversarial environments by using advanced decision-making and navigation algorithms, breakthrough materials and structures and intelligent power systems. The ARC is at the heart of an ecosystem of research with experts in multiple disciplines who address scientific challenges to enable and transfer advanced capabilities to DoD at the speed of relevance. In particular, the ARC is the flagship academic partner of the U.S. Army Ground Vehicle Systems Center, and leads the way in the revolutionary change toward autonomy in commercial and military ground systems.

Bio: Bogdan I. Epureanu is the Roger L. McCarthy Professor and Arthur F. Thurnau Professor in the Department of Mechanical Engineering at the University of Michigan and has a courtesy appointment in electrical engineering and computer Science. He received his Ph.D. from Duke University in 1999. He is the director of the Automotive Research Center, which leads the way in areas of autonomy of ground systems, including vehicle dynamics, control and autonomous behavior, human-autonomy teaming, high performance structures and materials, intelligent power systems and fleet operations and vehicle system of systems integration.

His research focuses on nonlinear dynamics of complex systems, such as teaming of autonomous vehicles, enhanced aircraft safety and performance, early detection of neurodegenerative diseases, and forecasting tipping points in engineered and physical systems such as disease epidemics and ecology. His research brings together interdisciplinary teams and consortia such as government (NIH, NSF, DOE, DOD), industry (Ford, Pratt & Whitney, GE, Airbus) and academia. He has published over 350 articles in journals, conferences and books. He is the Editor in Chief of the ASME Journal of Computational and Nonlinear Dynamics and serves on several other editorial boards.