An Engineer's Foray into Topological Learning: Addressing Challenges in Mobile Robot Perception

Abstract: Topological learning (TL), referring to a synergy of computational topology and machine learning, has recently emerged as an effective pattern recognition framework for noisy, high-dimensional problems. The recognition happens by first identifying the topological structures that encode the shape and connectedness information among the observations (samples), and then characterizing the structures based on their relative persistence over a wide range of spatial and/or temporal scales. In this talk, I will discuss successful demonstrations of TL for challenging mobile robot visual perception problems in unseen, cluttered environments. Our novel adaptation of TL recognizes occluded objects significantly more accurately than state-of-the-art shape or learning-based methods without requiring real-world training samples. I will conclude by pointing out ongoing and future research directions of TL for high-fidelity semantic mapping and active exploration of such environments.

Bio: Ashis G. Banerjee is an Associate Professor of Industrial & Systems Engineering and Mechanical Engineering at the University of Washington (UW). Prior to joining UW, he was a Research Scientist at GE Global Research and a Postdoctoral Associate at MIT. He obtained his Ph.D. and M.S. in Mechanical Engineering from the University of Maryland, College Park, and B.Tech. in Manufacturing Science and Engineering from IIT Kharagpur. He has broad research interests in autonomous robotics and cyber-physical systems. He is an elected Senior Member of the IEEE, and serves as a Senior Editor for IEEE Robotics and Automation Letters.