The concept of collective autonomous systems – e.g., robot swarms, fleets of aerial vehicles, and physical networks – promises task parallelism, resilience and operational flexibility in applications ranging from search and rescue, disaster response and critical infrastructure restoration to urban air mobility. “Adaptive coordination” of the constituent systems is critical for achieving generalizability, scalability, and fast decision-making, as an important step towards realizing this promise. Often these coordination problems, when mathematically represented, exhibit as notoriously challenging combinatorial optimization problems, direct solutions to which are usually intractable, time-prohibitive, and/or not suited for decentralized implementation. Graph abstraction of the associated task and/or agent space on the other hand a provides a natural way of expressing these problems; and enables the use of graph neural nets (GNNs) as decision-support policies. This seminar talk will touch upon our work on such a spectrum of graph learning methods, applied to solving operations planning problems in the following use cases: multi-drone disaster relief delivery, fleet planning of flying taxis, and restoration of disrupted power grids.
Souma Chowdhury is a Professor of Mechanical and Aerospace Engineering, with adjunct affiliation in Computer Science and Engineering, at University at Buffalo (UB). In addition, he serves as the Co-Director of both the Center for Embodied Autonomy and Robotics (CEAR) and the M.S. in Robotics program at UB. Dr. Chowdhury graduated with his Ph.D. in Mechanical Engineering from Rensselaer Polytechnic Institute (RPI) in Troy, NY in 2012. His research interests lie at the intersections of multi-fidelity optimization, physics-informed machine learning and graph learning, with applications to the design and operation of autonomous aerospace systems, robot swarms and resilient networks. He has co-authored 62 peer-reviewed journal articles, 138 conference proceedings articles, and 3 book chapters in related topics. His work has been recognized by multiple best paper and best student paper awards at IEEE, ACM and AIAA conferences. He is a Fellow of ASME, an Associate Fellow of AIAA, and a Senior Member of IEEE, and currently serves as the Conference Chair of the ASME Design Automation Conference.
Souma Chowdhury
Professor,
Department of Mechanical and Aerospace Engineering
Adjunct Associate Professor,
Department of Computer Science and Engineering
Co-Director,
Center for Embodied Autonomy and Robotics (CEAR)
Co-Director, Engineering Science MS in Robotics
School of Engineering and Applied Sciences
University at Buffalo