Covariant today announced the close of a $40 million series B funding round to bring its robotic control systems to additional industries and create more systems capable of picking, placing, and unloading objects in warehouses. Until now, Covariant has focused its efforts on ecommerce picking robots in highly automated warehouses. It may be best known for its work in robotic grasping, the task of picking up objects with a robotic hand or gripper.
The startup — whose founders met at OpenAI and the University of California, Berkeley — has raised $67 million to date. After emerging from stealth earlier this year with support from deep learning luminaries like Geoffrey Hinton, Jeff Dean, and Yann LeCun, Covariant stated that the Covariant Brain system is capable of picking and packing some 10,000 items with 99% accuracy.
Robotics manufacturer ABB signed a partnership with Covariant in February, following a picking and sorting test held by ABB last year in which Covariant outperformed 20 other systems. In March, Covariant and Knapp signed a partnership to release a picking robot solution.
Covariant has primarily deployed robots in warehouses with high levels of automation, but the funding will be used to expand the company’s footprint to include warehouse environments with low rates of automation, or where work is done entirely with human labor today.
Covariant CEO Peter Chen told VentureBeat examples of low automation industries include mail and parcel delivery, with companies like UPS or the U.S. Postal Service.
“There are a lot of tasks where grasping is the first step in robotic manipulation, but it’s one of the many steps in other use cases that we’re looking into tackling — and that obviously go beyond just the logistics supply chain industry, like going to manufacturing, recycling, agriculture. These are places where people still use their hands a lot to do very repetitive kinds of tasks,” Chen told VentureBeat in a phone interview.
A series of studies MIT economists released this week found that robotics are most prevalent in four manufacturing industries: automakers (38% of robots in use), electronics (15%), plastics and chemicals (10%), and metals manufacturers (7%). The study also found that robots replace on average 3.3 jobs, but businesses that move quickly to adopt robots can also add employees to their payroll.
Since learning that the core principles behind sorting ecommerce items in warehouses apply to other industrial applications, Chen expects Covariant will begin to develop robotic systems that go beyond tasks like loading and unloading boxes.
“Even though we have seen our robots operating in high-automation warehouses doing order picking and packing orders for consumers, the underlying technology is a lot more extensible than that, and that’s the key thing that we look to bring more to markets with our partners and solve more additional use cases,” Chen said.
He said Covariant has recently seen increased usage from clients hoping to avoid supply chain disruption. Since the start of the pandemic, Chen said, clients want robots for consistency and reliability or to avoid a slowdown in case of shelter-in-place orders in the future.
“What COVID-19 has shown us is some of the vulnerabilities and weaknesses in the supply chain, and now [the question is] ‘How can we invest in the next generation of robotics to help us be more resilient?'” he said.
In addition to opening up new industries, the funding will also be used to grow the research and engineering teams for the Covariant Brain robotics system.
The round was led by Index Ventures, with participation from Amplify Partners and Radical Ventures. Index Ventures partner Mike Volpi will join Covariant’s board of directors.
Covariant was founded in September 2017 and is based in Berkeley, California, with plans to move to nearby Emeryville in the weeks ahead.
In other news, last week Covariant cofounder and Berkeley AI Research codirector Pieter Abbeel open-sourced RAD, a module the team says is capable of improving any reinforcement learning algorithm, and published other reinforcement learning work at the ICLR machine learning research conference.