Planning and Resilient Execution of Policies For Manipulation in Contact\n with Actuation Uncertainty

Abstract

We propose a method for planning motion for robots with actuation uncertainty\nthat incorporates contact with the environment and the compliance of the robot\nto reliably perform manipulation tasks. Our approach consists of two stages:\n(1) Generating partial policies using a sampling-based motion planner that uses\nparticle-based models of uncertainty and simulation of contact and compliance;\nand (2) Resilient execution that updates the planned policies to account for\nunexpected behavior in execution which may arise from model or environment\ninaccuracy. We have tested our planner and policy execution in simulated SE(2)\nand SE(3) environments and Baxter robot. We show that our methods efficiently\ngenerate policies to perform manipulation tasks involving significant contact\nand compare against several simpler methods. Additionally, we show that our\npolicy adaptation is resilient to significant changes during execution; e.g.\nadding a new obstacle to the environment.\n