Planning and Resilient Execution of Policies For Manipulation in Contact with Actuation Uncertainty
This addresses manipulation reliability for robots in uncertain environments, though it appears incremental as it builds on existing sampling-based planning and adaptation techniques.
The authors tackled the problem of robot manipulation under actuation uncertainty by developing a two-stage method that plans motion with contact and compliance models, then adapts policies during execution. They demonstrated efficient policy generation for manipulation tasks with significant contact and resilience to environmental changes like new obstacles.
We propose a method for planning motion for robots with actuation uncertainty that incorporates contact with the environment and the compliance of the robot to reliably perform manipulation tasks. Our approach consists of two stages: (1) Generating partial policies using a sampling-based motion planner that uses particle-based models of uncertainty and simulation of contact and compliance; and (2) Resilient execution that updates the planned policies to account for unexpected behavior in execution which may arise from model or environment inaccuracy. We have tested our planner and policy execution in simulated SE(2) and SE(3) environments and Baxter robot. We show that our methods efficiently generate policies to perform manipulation tasks involving significant contact and compare against several simpler methods. Additionally, we show that our policy adaptation is resilient to significant changes during execution; e.g. adding a new obstacle to the environment.