How deep-machine learning is being used to meet the growing demand for package handling and product shipments.
Asked to choose a superpower, few people would think “suction.” But it turns out that robots with suction hands can achieve superhuman sorting performance, a capability that could soon revolutionize e-commerce warehouses.
In March, Ambi Robotics, a company co-founded in 2019 by Berkeley engineering professor and roboticist Ken Goldberg and four graduate students, announced two flagship products. The first, AmbiSort, is a technology that enables industrial robots to “grasp, scan, and place objects at twice the speed of human workers,” Goldberg said. The second, AmbiKit, is a robotic system for assembling subscription boxes, gift sets, and the like.
The brains of the suction-equipped robots are based on the Dexterity Network (Dex-Net), a deep machine learning program created by Berkeley researchers to train robots to grasp a vast array of novel objects—a notoriously challenging task for machines.
Ambi Robotics raised $6.1 million in seed funding and aims to use the technology to address the growing demand for package handling due to the increase in online shopping, a trend that has accelerated during the pandemic.
As Goldberg sees it, “E-commerce demand will continue to grow, and robots will be there to fill the gaps when there are not enough human workers to do the job—and to tackle the dangerous, dull, and dirty jobs so that human workers can focus on what they do best.”
From the Summer 2021 issue of California.
