Learning Robotic Manipulation of Granular Media

Abstract

In this paper, we examine the problem of robotic manipulation of granular media. We evaluate multiple predictive models used to infer the dynamics of scooping and dumping actions. These models are evaluated on a task that involves manipulating the media in order to deform it into a desired shape. Our best performing model is based on a highly-tailored convolutional network architecture with domain-specific optimizations, which we show accurately models the physical interaction of the robotic scoop with the underlying media. We empirically demonstrate that explicitly predicting physical mechanics results in a policy that out-performs both a hand-crafted dynamics baseline, and a "value-network", which must otherwise implicitly predict the same mechanics in order to produce accurate value estimates.

arXiv commentProceedings of the Conference on Robot Learning 2017 (CoRL) (to appear)

References in corpus (7)

• Playing Atari with Deep Reinforcement Learning
• A Terradynamics of Legged Locomotion on Granular Media
• Learning to Navigate in Complex Environments
• Accelerating Eulerian Fluid Simulation With Convolutional Networks
• Sensitive dependence of the motion of a legged robot on granular media
• A Compositional Object-Based Approach to Learning Physical Dynamics
• The effect of limb kinematics on the speed of a legged robot on granular media

Cited by in corpus (6)

• A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms
• Manipulation of Granular Materials by Learning Particle Interactions
• Target-mass Grasping of Entangled Food using Pre-grasping & Post-grasping
• A Survey of Behavior Learning Applications in Robotics -- State of the Art and Perspectives
• Learning compositional models of robot skills for task and motion planning
• Active model learning and diverse action sampling for task and motion planning