Distributed Inference for Robotic Perception and Planning in Unstructured Environments
Abstract: To achieve the long-promised general robot assistants capable of operating in our uncertain and unstructured world, we need efficient, reliable, and cooperative AI systems. My research explores advancements in distributed methods for efficient, adaptive, and robust inference for robotic perception and planning problems. Distributed inference leverages graphical models to break up complex problems into multiple smaller subproblems which can be solved in parallel. In this talk, I will present my work on distributed inference for challenging robotic problems including robot perception and multi-robot coordination. I will demonstrate how nonparametric representations of uncertainty help tackle high-dimensional problems with multiple diverse solutions efficiently. First, I will present my contributions to nonparametric belief propagation for the problem of articulated object localization in cluttered scenes towards robot manipulation of hand-tools. Next, I will show how these ideas are extended to distributed planning over multi-robot teams through a new algorithm called Stein Variational Belief Propagation. In the last part of the talk, I will show how lessons on distributed inference can be applied to robotics education to meet the growing demand for robotics and AI curricula across universities.
Bio: Jana Pavlasek is a PhD candidate in the Robotics Department at the University of Michigan advised by Professor Chad Jenkins. Her research interests include robotic perception and planning under uncertainty in challenging, real-world environments. Particularly, her work focuses on scalable distributed inference algorithms, including perception in clutter and multi-robot coordination. Jana is the developer and instructor for a core course on introductory robotics for the Robotics undergraduate degree at the University of Michigan, which has also been taught across several other universities as part of a distributed teaching initiative. Jana holds a Masters in Robotics from the University of Michigan. Prior to joining Michigan, Jana received a Bachelor’s degree in Electrical Engineering (Honours) at McGill University. She is the recipient of an NSERC Postgraduate Scholarship and has interned at NVIDIA Research.
Location: Virtual – Zoom Only
Zoom Link: https://wpi.zoom.us/j/92946413217