In 2012, Amazon quietly acquired a robotics startup called Kiva Systems, a purchase that dramatically improved the efficiency of the company’s e-commerce operations and launched a broader revolution in warehouse automation.
Last week, the e-commerce giant announced another potentially equally significant deal, agreeing to hire the founder of Covariant, a startup that has been testing ways to use AI to further automate the pickup and handling of a variety of objects.
Covariant may have found it difficult to commercialize its AI-powered industrial robots given the high costs and fierce competition. The deal, which will also see Amazon license Covariant’s models and data, could usher in a new revolution in e-commerce. Given Amazon’s massive scale and vast amounts of data, it may be difficult for competitors to match.
The deal also marks an example of a big tech company acquiring an AI startup’s core talent and expertise without acquiring it: Amazon struck a similar deal with startup Adept in June, Microsoft struck a deal with Inflection in March, and Google hired the founder of Character AI in August.
In the 2000s, Kiva developed a way to move goods around warehouses by using chunky robots to lift shelves of items and carry them to human pickers, freeing workers from having to walk miles each day to find different products. Kiva’s mobile robots are similar to those used in manufacturing, and the company uses clever algorithms to coordinate the movements of thousands of robots in the same physical space.
Amazon’s army of mobile robots has grown from about 10,000 in 2013 to 750,000 by 2023, and the company’s scale allows it to deliver millions of items faster and cheaper than any other company.
As WIRED revealed last year, Amazon has been developing new robotic systems in recent years that use machine learning to recognize, grab, and sort packed boxes, among other tasks. Again, Amazon is using its scale to its advantage, collecting training data as items move through its facilities to help improve the performance of its various algorithms. This effort has already enabled further automation of tasks previously performed by human workers in some of its fulfillment centers.
But one task remains stubbornly difficult to mechanize: physically gripping items. This requires adaptive thinking to account for friction, slippage, and other issues, and robots will inevitably be faced with unfamiliar and tricky items in Amazon’s vast inventory.
Covariant has spent the past few years developing AI algorithms with more generalized capabilities to more reliably process a variety of items. The company was founded in 2020 by Peter Abeel, a professor at the University of California, Berkeley who pioneered the application of machine learning to robotics, and several of his students, including Peter Chen, who became the company’s CEO, and Rocky Duan, the company’s CTO. With this week’s deal, all three will join Amazon, along with several of the startup’s research scientists.
“Covariant’s model will be part of the robotics operations system across our fulfillment network,” Amazon spokesperson Alexandra Miller told WIRED. The tech giant declined to disclose financial details of the deal.
Abeer was an early employee at OpenAI, and his company is inspired by the success story of ChatGPT. In March, Covariant showed off a chat interface for robots and announced it had developed a basic model of robotic grasping. This is
