Enhancing Robot Perception for Robust Autonomy in Unstructured Environments
Abstract: Perception and action are deeply intertwined in humans; robustly replicating the perception-action coupling is critical to building `generalist' robots and achieving human-like performance in them. The ability to capture statistical correlations between data makes deep learning the method of choice for perception in structured environments. However, unstructured environments necessitate methods that go beyond statistical correlation. In this talk, I will present my work on visual object recognition in unseen environments, where I enhance the performance of learning-based classifiers by combining them with topological object representations that embody human-like reasoning. This novel approach recognizes occluded objects with substantially higher accuracy than other state-of-the-art methods without requiring real-world training data. I will also briefly describe my work on designing perception methods for a broader range of systems, such as manufacturing inspection systems, autonomous vehicles, human ergonomic risk assessment systems, and optical tweezers for microscale manipulation. I will conclude by pointing out future research directions to achieve perception-action robustness.
Bio-sketch: Ekta U. Samani is a Postdoctoral Scientist at Amazon. She obtained her Ph.D. in Mechanical Engineering from the University of Washington (UW), Seattle. She also obtained an M.S. in Mechanical Engineering: Data Science from UW and a B.Tech. in Electrical Engineering with a minor in Computer Science and Engineering from the Indian Institute of Technology (IIT), Gandhinagar. Her research focuses on combining model-based reasoning with advanced pattern recognition for robust perception in autonomous systems. She was a 2023 Robotics: Science and Systems (RSS) Pioneer, received the UW Women Engineers Rise Outstanding Student Award, and won the second-place prize in the UW Three Minute Thesis (3MT) competition. She serves as an associate editor for the IEEE Robotics and Automation Letters.
Zoom Link: https://wpi.zoom.us/j/92946413217