H.C. Torng Lecture Series in Computer Engineering: Nils Napp of University at Buffalo

Description

Material Modeling and Control for Physical Distributed Systems Many of the models used to understand emergent swarm behaviors abstract away physics as much as possible for mathematical convenience. However, some of the most interesting things in happen when minute physical interactions result in complex and robust behavior, for example during self-assembly or in cooperative nest building. In this talk I will present research on modeling, building, and controlling distributed systems that create physical artifacts at a variety of scales and complexity. One particular application is modeling teams of agents that can modify a shared, cluttered environment as a way to automate construction in remote locations and disaster sites. The main difficulty is in representing a complex environment and an agent’s actions in a tractable way. The key result is a model and algorithm for construction with amorphous and irregularly shaped materials. The resulting system can robustly build structures that adapt to their environment and would be useful, for example, to build emergency levees using sandbags. Each agent essentially performs feedback control on the structure to achieve a specific, locally-measurable structural functionality. More generally, I will present methods for both open and closed loop control of such distributed systems. Nils Napp is an Assistant Professor of Computer Science and Engineering at the University at Buffalo since 2014. Previously he was a graduate student at the University of Washington (Ph.D., EE 2011) and a postdoc at Harvard's Wyss Institute (2011-14). His research focuses on algorithms at the intersection of collaborative behavior and physical embodiment. When taking inspiration from natural systems, the central question is why they work so well, and if those insights can be applied to engineered systems. Since machines outperform biology in many types of communication and computation, the resulting systems often look different from either their biological inspiration or traditional robots.