Exploiting Contact for Efficient Motion Planning Under Uncertainty

In this paper we want to argue for contact as an enabler for efficient motion planning under uncertainty. Controlled and desired contact can project a high dimensional belief state to a lower-dimensional manifold. A robot can sequence these projections to reduce uncertainty about its state. For realistic applications, these uncertainty reducing actions must be sequenced with uncertainty-increasing free space motion. We present a sampling-based motion planner that searches a belief state over configurations augmented with contact information. The planner finds robust contact-exploiting policies under significant uncertainty in robot and world model. We validate these policies on a seven-DoF robot manipulator in simulation and real world experiments.