When: 21nd of April, 1:00pm AEDT
Where: This seminar will be partially presented at the ACFR seminar area, J04 lvl 2 (Rose St Building) and partially online via Zoom. RSVP
Speaker: Dr. K. M. Brian Lee
Title: Beyond Binary Collision Checking for Robot Safety: Spatial Representations, Free-Space Bubbles, and Contact-Aware Planning
Abstract:
A usual assumption in robot motion planning is that safety reduces to binary collision checking: a point in configuration space is either in collision or not. In this talk, I argue that extending binary collision checking to richer geometric notions of safety is advantageous for efficient and contact-aware motion planning. I will first discuss a family of learned spatial representations that provide the dense, queryable geometric structure that downstream planners need. I will then show how a signed distance field learned from streaming point cloud data enables motion planning on a graph of free-space bubbles, which reduces collision checks, supports continuous trajectory optimization, and provides runtime safety guarantees. Finally, I will present ongoing work on contact-aware planning approaches where environmental contact is treated as a resource rather than a failure mode, with applications to non-prehensile manipulation and continuum robots. I will close with a discussion of open directions, including further generalizations of safety notions.
Bio:
K. M. Brian Lee is a Postdoctoral Scholar at the Contextual Robotics Institute, UC San Diego. He received the Ph.D. degree in Robotics from the University of Technology Sydney and the B.Eng. (Hons I) degree in Mechatronics (Space) from the University of Sydney. He was recognized as an RSS Pioneer of 2023, and was the recipient of the UTS Research Excellence Scholarship.