In this talk I will present an overview of my recent work developing autonomous motion planning algorithms with rigorous guarantees on their safety and optimality. The basic idea underpinning much of this work is sampling-based motion planning, which discretizes a robot's obstacle-free configuration space and finds a path through a locally-connected graph on this discretization. This class of algorithms has had great success planning when the configuration space dimension (the robot's degrees of freedom) is more than two or three. My work includes developing algorithms that reduce computational complexity relative to the state-of-the-art while maintaining asymptotic optimality, proving the first convergence rates for such algorithms, extending them to robots with differential constraints, and to settings with uncertainty in the environment or limited sensing radius. After overviewing the basic problem, approach, and my contributions, I will end by posing what I see as the next big problem in this domain: quantifying uncertainty while reinforcement learning to allow for uncertain robot dynamics.
Lucas Janson is an Assistant Professor in the Department of Statistics at Harvard University, where he works on high-dimensional inference, autonomous robotic motion planning, and statistical machine learning. Prior to Harvard, he was a PhD student in Statistics at Stanford University advised by Professor Emmanuel Candès.