Robotics maker ABB has partnered with A.I. startup Covariant

Industrial robotics maker ABB is partnering with A.I. startup Covariant to create robots that can better pick and grasp thousands of different kinds of objects in warehouses.

ABB is among the largest makers of robots for automakers, but it wants to expand into other industries. One promising area is logistics, in which retailers like Amazon use cutting-edge technologies in warehouses to package and ship orders.

Sami Atiya, ABB’s president of robotics and discrete automation, explained that while ABB uses A.I. technologies like computer vision to help its robots understand whether a particular car part has been welded correctly, it wanted outside help to create robots that can pick and grasp objects. Although it may seem like a simple task, technologists have long sought to create robotic arms that can handle objects with human-like dexterity.

To evaluate potential partners, Switzerland-based ABB created a contest in which 10 European and 10 U.S. companies tested their robotic gripping software on a variety of items. Covariant, founded by robotics researchers from the University of California at Berkeley, and the research lab OpenAI, won the contest based on it being the only company whose software could recognize a number of different items and needed no human intervention, Atiya said.

Covariant’s robotic gripping software is based on the A.I. technique of reinforcement learning, in which software learns to pick and place objects through repetition. The combination of reinforcement learning and neural networks—software that learns and adapts from vast quantities of information—have led to recent breakthroughs including robots learning to pick up objects they’re unfamiliar with without breaking or misplacing those objects.

Peter Chen, Covariant’s CEO, said as more established robot companies like ABB enter new markets, they want to create robots that can adapt better than the robots that they typically sell to carmakers. Those robots do a number of repetitive tasks like lifting up automobile frames, installing windshields, and mounting wheels—tasks that typically require distinct parts that the robots have been specifically built to handle.

“The logistics space is kind of nascent,” said Chen. “The fact that ABB needed to run this global competition tells you something—it tells you there is not a lot of knowledge in this space.”

By partnering with Covariant, ABB will get access to more customers as well as learn about some of the technology in ABB’s existing robots that could help Covariant better embed its A.I. software for customers.

The partnership underscores how major robotics companies are turning to startups for help with creating newer technologies. For instance, Japan-based industrial equipment maker IHI Corporation recently partnered with the A.I. startup Osaro on a similar initiative related to robots picking up and grasping objects. Meanwhile, another Japanese robotics giant, Fanuc, recently partnered with the startup Soft Robotics to use the startup’s gripping systems in its own robotic systems.

Said Chen, “What Silicon Valley companies are good at is building the software —we need partners like ABB.”