Accelerating Grasp Exploration by Leveraging Learned Priors

TL;DR

This paper introduces a Thompson sampling algorithm that uses learned priors to improve robotic grasp exploration on novel objects, significantly increasing success rates and efficiency.

Contribution

The novel approach combines learned priors with online exploration to enhance grasping of complex and unseen objects, outperforming baseline methods.

Findings

64.5% higher average total reward than greedy baseline

Within 5.7% of oracle baseline performance

Effective on 3000 object poses across 300,000 runs

Abstract

The ability of robots to grasp novel objects has industry applications in e-commerce order fulfillment and home service. Data-driven grasping policies have achieved success in learning general strategies for grasping arbitrary objects. However, these approaches can fail to grasp objects which have complex geometry or are significantly outside of the training distribution. We present a Thompson sampling algorithm that learns to grasp a given object with unknown geometry using online experience. The algorithm leverages learned priors from the Dexterity Network robot grasp planner to guide grasp exploration and provide probabilistic estimates of grasp success for each stable pose of the novel object. We find that seeding the policy with the Dex-Net prior allows it to more efficiently find robust grasps on these objects. Experiments suggest that the best learned policy attains an average total reward 64.5% higher than a greedy baseline and achieves within 5.7% of an oracle baseline when evaluated over 300,000 training runs across a set of 3000 object poses.