In 2012, Amazon quietly acquired a robotics startup called Kiva Systemswhich significantly improved the efficiency of e-commerce operations and started a broader revolution in warehouse automation.
Last week, the e-commerce giant announced another deal that could prove just as significant, agreeing to hire covariant foundersa startup testing ways to apply artificial intelligence to automate the processes of lifting and moving a wide range of physical objects.
Covariant may have struggled to commercialize its AI-powered industrial robots due to high costs and stiff competition. The deal, which also allows Amazon to license Covariant’s models and data, could usher in another e-commerce revolution—one that competitors will struggle to match given Amazon’s enormous scale and data assets.
The deal is also an example of a Gigantic Tech company acquiring key talent and expertise from an AI startup without actually buying the company. Amazon came similar agreement with the startup Adept in June. In March, Microsoft hit Dealing with Inflectionand in August Google hired the founders of Character AI.
In the 2000s, Kiva developed a way to move products through warehouses by using crouched robots to lift and carry full shelves to human pickers—a trick that meant workers no longer had to walk miles each day to find items. Kiva’s mobile bots were similar to those used in manufacturing, and the company used clever algorithms to coordinate the movement of thousands of bots in the same physical space.
Amazon’s army of mobile robots grew from around 10,000 in 2013 to 750,000 in 2023, and the sheer scale of the company’s operations meant it could deliver millions of items faster and cheaper than anyone else.
As WIRED revealed last year, Amazon has been developing fresh robotic systems in recent years that rely on machine learning to perform tasks like perceiving, grasping, and sorting packed boxes. Again, Amazon is using scale to its advantage, with training data being collected as items flow through its facilities, helping to improve the performance of various algorithms. This effort has already led to further automation of work that was previously performed by workers in some fulfillment centers.
However, one task that remains stubbornly tough to mechanize is the physical handling of products. This requires adaptability to account for things like friction and slippage, and the robots will inevitably have to deal with unfamiliar and awkward items among Amazon’s enormous inventory.
Covariant has spent the last few years developing AI algorithms with a more general ability to handle a range of items more reliably. The company was founded in 2020 by Pieter Abbeelprofessor at the University of California at Berkeley who did pioneering work on the application of machine learning to roboticsalong with several of his students, including Peter Chen, who became Covariant’s CEO, and Rocky Duane, the company’s CTO. As part of the week-long agreement, all three, along with several research scientists from the startup, will join Amazon.
“The Covariant models will be used to power some of the robotic manipulation systems in our fulfillment network,” Amazon spokeswoman Alexandra Miller told WIRED. The tech giant declined to disclose financial details of the deal.
Abbeel was an early OpenAI employee, and his company drew inspiration from the success story of ChatGPT. In March, Covariant demonstrated a chat interface for its robot and said it had developed a baseline model for robotic grasping, meaning an algorithm designed to become