Technology developed by Berkeley robotics researchers put to real world use
Covariant.AI, a robotics startup founded by researchers and alumni at Berkeley Engineering, is starting to see the fruits of its labor.
The company's robots use reinforcement learning to master the task of sorting through thousands of electrical parts that pass each day on the conveyer belt at the Obeta warehouse in Ludwigsfelde, Germany. In the past, picking through a bin of random items would be too complex of a task for robots. A human touch has been needed.
Until now. Covariant has created robots that can independently learn new tasks through trial and error rather than tediously teaching the robots to recognize each single part under an endless range of conditions and variables. Shapes vary, as do surfaces. One light switch might be upside down, the other right-side up. The next electrical gadget might be in a plastic bag that reflects light in ways a robot has never seen.
Covariant was founded by Pieter Abbeel, a professor in the Department of Electrical Engineering and Computer Sciences, along with Berkeley graduates Peter Chen (B.A.'05 Math), Rocky Duan (Ph.D.'17 CS) and Tianhao Zhang (M.S.'18 CS).