E-commerce continues to expand and reach new levels during recent holidays. To quickly fulfill the huge volume and variety of orders, companies such as Amazon, Walmart and Alibaba are investing heavily in new warehouses. To address the shortage of workers, many companies are considering robots. However, the reliable grasp of a wide range of products remains a major challenge for robotics.
In a paper published on Wednesday, January 16, in Robotic science, engineers at the University of California, Berkeley present a "ambidextrous" approach to understanding a wide range of forms of unclassified objects.
"Any gripping device can not handle all the objects," said Jeff Mahler, postdoctoral researcher at UC Berkeley and lead author of the paper. "For example, a suction cup can not create a seal on porous objects such as clothing and parallel jaw gripping devices may not be able to reach both sides of tools and toys."
Mahler works in the laboratory of Ken Goldberg, Professor UC Berkeley, with joint appointments in the Department of Electrical Engineering and Computer Science and the Department of Industrial Engineering and Operations Research.
The robotic systems used in most eCommerce centers are based on suction pliers that can limit the range of objects they can understand. UC Berkeley paper introduces an "ambidextro" approach that is compatible with a variety of pliers. The approach is based on a common reward function for each type of gripper that quantifies the likelihood that each gripper will succeed. This allows the system to quickly decide which gripper to use for each situation. To efficiently calculate a reward function for each type of gripping device, the paper describes a process of learning reward functions by training on large sets of rapidly generated synthetic data using structured domain randomizations and analytical sensing and physics and geometry models of each clamping device.
When researchers trained rewarding functions for a parallel jaw clamping device and a vacuum cleaner attachment on a two-arm robot, they discovered that their system had cleaned up to 25 previously unseen objects at a rate of over 300 drops per hour with 95% reliability.
"When you are in a warehouse that puts the packages together for delivery, the items vary considerably," Goldberg said. "We need a variety of gripping devices to solve a variety of objects."
This soft robotic fastener can screw you in bulbs
J. Mahler et al., "Learning Ambidextrous Robot Policies to Understand" Robotic science (2018). robotics.sciencemag.org/lookup … /scirobotics.aau4984
University of California – Berkeley
"Ambidextro Robots" could dramatically accelerate electronic commerce (2019, January 16)
taken over January 17, 2019
This document is subject to copyright. Apart from any proper trading for private study or research, no
a part may be reproduced without written permission. Content is provided for informational purposes only.