For the last several years, the Delivering the Future event has showcased the latest technologies powering Amazon operations. Seattle’s 2023 event showcased updates to the company’s pharmacy offerings and drone deliveries.
This year in Nashville, Amazon discussed AI updates to its shopping experiences and how the company is using computer vision to further trim package delivery time. While the two-day event didn’t find the company showcasing any new robotics systems, it did offer some key insight into how it’s integrating existing offerings.
Shortly after the event, TechCrunch sat down with Amazon Robotics chief technologist Tye Brady. The conversation has been an annual event for the last few years, giving us an opportunity to dig deeper into how Amazon’s robotics story has changed in the past 12 months and how the next 12 will evolve.
According to Amazon’s internal figures, the company currently has 750,000+ robots deployed in its U.S. fulfillment centers. It’s the same public figure the company touted in 2023. This isn’t the whole story, however. The 750,000 figure only encompasses the company’s autonomous mobile robots (AMRs).
These are the familiar wheeled systems that have patrolled Amazon warehouse floors since the company acquired Kiva Systems back in 2012. These tote robots, which the company also refers to as its drive-train systems, include a number of different models, including the autonomous Proteus system, which was unveiled at Re:Mars back in 2022.
ARMs form the vast majority of Amazon’s fleet, but other form factors have carved out their own space on the floor. The next largest category is Amazon’s robotic arms, which now include Robin, Cardinal, and Sparrow, each tasked with sorting and stacking objects.
The most recent entry into the Amazon Robotics family is Sequoia, unveiled at the 2023 Delivering the Future event. The name, borrowed from the massive redwoods of Northern California, may well be an homage to the system’s size and scope. Sequoia is an automated storage and retrieval system similar in principle to those offered by firms like AutoStore.
The first Sequoia system went online in 2023 at a Houston-area fulfillment center. On Wednesday, Amazon announced that a system 5x its size now forms the heart of a massive warehouse in Shreveport, Louisiana. The fulfillment center itself isn’t new, but it is being dramatically expanded to span in excess of 3 million square feet.
Amazon certainly possesses the resources to build new greenfield robotic fulfillment centers from the ground up. Instead, the massive retailer is focused on retrofitting existing brownfield warehouses. While less resource intensive, this method forces the company to work around existing delivery operations of “fix the airplane while it’s flying,” as Brady puts it.
The Shreveport center, the first of Amazon’s “Gen 12” buildings, will ultimately utilize 10x the number of robots as their predecessors — a figure the company has not yet disclosed. Brady adds that, along with these new robots come more robot-centric jobs. That means 25% more RME (reliability maintenance engineering) roles than before.
The company adds that, once Shreveport’s 55 football fields’ worth of fulfillment operations is up and running, it will employ 2,500 humans. Automation fans will tell you these technologies allow humans to focus on the things they can do that robots can’t. Brady is a big proponent of the notion.
Asked what jobs humans are still better suited for, he answers, “Problem solving, common sense, thinking with reason, understanding the big picture, understanding the context. Some physical tasks, as well.”
While Agility’s Digit robot had a star turn at the 2023 event, Amazon didn’t mention much on the humanoid front. Certainly, the company has been exploring the roles bipedal robots can play in its fulfillment centers, including a pilot with Agility announced last year. Things, however, have been quiet since the pilot’s completion.
“We’re still learning,” Brady says of the Agility partnership. “It’s slow and steady. ‘R&D’ is the best way I can capture it.” The slowness, he explains, is a product of finding ways such technology could slot into existing workflows.
“We start with the problem we are trying to solve,” Brady says. “When you have a piece of technology and say, ‘Hey, how do I apply this?’ that’s a dangerous path when you try to force things. The fact is, in our fulfillment centers, we have a lot of nice, poured concrete floors. Wheels are pretty good. But we also have stairs and uneven terrain when we start to go outside.
Brady confirmed that the partnership is still active but had no additional information to share with TechCrunch.
One partnership that’s grown much clearer in recent months is the UC Berkeley spinoff Covariant. In August, Amazon announced that it had hired the startup’s founders, Pieter Abbeel, Peter Chen, and Rocky Duan, along with around 25% of its employees. The move is a bid to expand the role of foundational models in the industrial setting.
Amazon notes, for instance, that its robot arm, Sparrow, “can now handle over 200 million unique products of all different shapes, sizes, and weights.” But there are always edge cases. Handling these will come down to both human employees and better-trained AI systems. Covariant, which operates in these sorts of massive datasets to fine-tune things like product pick and placement, will play a key role.
“We’re up and going and starting to work on some really meaty, very applied problems for machine learning,” Brady says of the deal.