We focus on learning to compute near-optimal plans which leverage environmental contact to mitigate action uncertainty, in hopes of enabling inexpensive robotic manipulators to perform precise assembly tasks.

Path planning classically focuses on avoid obstacles and environmental contact. However, some assembly tasks permit contact through compliance, and such contact allows for significantly better solutions under action uncertainty. Optimal manipulation plans which leverage environmental contact are difficult to compute. Environmental contact produces complex kinematics, rarely available in analytic form, and difficult to work with as a whole. Discretization over state and action space is usually unavoidable, and quickly becomes computationally intractable. We aim to investigate various methods to address this issue based on computational geometry and differential geometry planning.

Research Areas