In Short:
In 2012, Amazon acquired Kiva Systems, enhancing its warehouse efficiency. Recently, it hired Covariant’s founders, aiming to improve AI-driven automation for picking and handling products. This partnership allows Amazon to access Covariant’s AI models, potentially revolutionizing ecommerce. Unlike prior acquisitions, this deal focuses on gaining expert talent. Covariant’s technology aims to tackle the tough challenge of robotic product grasping.
Amazon made a significant move in 2012 by acquiring Kiva Systems, a robotics startup that subsequently transformed its ecommerce operations and initiated a broader revolution in warehouse automation.
Recently, the ecommerce giant announced another impactful agreement, this time agreeing to hire the founders of Covariant. This startup has been developing AI solutions aimed at automating the picking and handling of an extensive range of physical items.
The journey to commercializing AI-infused industrial robots has faced challenges, particularly regarding high costs and intense competition. However, this agreement, which also includes licensing Covariant’s models and data, could pave the way for another substantial evolution in ecommerce, one that may be difficult for competitors to replicate due to Amazon’s expansive operational scale and comprehensive data resources.
Strategic Talent Acquisition
This deal serves as a prime example of a Big Tech company’s strategy to acquire essential talent and expertise from an AI startup without opting for a full acquisition. In a similar vein, in June, Amazon reached an agreement with Adept, and earlier this year, Microsoft forged a deal with Inflection. In August, Google hired the founders of Character AI.
Amazon’s Robotic Innovations
In the early 2000s, Kiva pioneered a method for transporting products through warehouses using small robots that could lift and move stocked shelves to human pickers, significantly reducing the distance workers had to cover daily. Kiva’s mobile robots employed sophisticated algorithms to orchestrate the movements of thousands of bots within limited spaces.
Since its acquisition, Amazon’s fleet of mobile robots has escalated from roughly 10,000 in 2013 to an impressive 750,000 by 2023. This extraordinary scale enables the company to deliver millions of items more quickly and cost-effectively than its competitors.
As reported by WIRED last year, Amazon has developed advanced robotic systems that utilize machine learning for tasks such as perceiving, grasping, and sorting boxes. Amazon’s vast scale allows it to gather training data as items move through its facilities, subsequently enhancing the performance of various algorithms. This initiative has already resulted in greater automation of tasks previously handled by human laborers at some fulfillment centers.
Challenges Ahead
Despite these advancements, the challenge of mechanizing the physical handling of products remains particularly demanding. Successfully grasping items involves addressing variables like friction and slippage, and robots must also contend with an array of unfamiliar and awkward items within Amazon’s extensive inventory.
Covariant has dedicated recent years to the creation of AI algorithms capable of handling diverse items with greater reliability. Founded in 2020 by Pieter Abbeel, a professor at the University of California, Berkeley, alongside several of his students—including Peter Chen, the CEO of Covariant, and Rocky Duan, the CTO—the startup’s approach draws inspiration from the success of ChatGPT.
This week’s agreement will bring all three founders, along with numerous research scientists from Covariant, into Amazon’s fold. An Amazon spokesperson, Alexandra Miller, stated that “Covariant’s models will be used to power some of the robotic manipulation systems across our fulfillment network.” The details of the deal’s financial arrangements have not been disclosed.
