Global Search with Bernoulli Alternation Kernel for Task-oriented Grasping Informed by Simulation
Researchers combine simulated grasp scoring with a Bernoulli Alternation Kernel to adapt task-oriented grasps online on an ABB YuMi.
Why it matters
We develop an approach that benefits from large simulated datasets and takes full advantage of the limited online data that is most relevant. We propose a variant of Bayesian optimization that alternates between using informed and uninformed kernels.
Evidence notes
- Hardware experiments use an ABB YuMi, a rigid parallel gripper and Microsoft Kinect point clouds to test novel objects across six tasks.
- The simulated training set covers 605 object meshes from 13 categories and 4,500 grasps per object.
- Bayesian optimization found successful task-appropriate grasps for the spatula and mug handover tasks on its fourth trial and the pan task on its fifth.
Company context
ABB is a publicly listed electrification and automation group. It currently owns ABB Robotics, whose signed sale to SoftBank remains pending, while Machine Automation continues inside ABB's Automation business.