The turn of the millennium marked a pivotal era for retail, as the Internet sparked an unprecedented surge in e-commerce. Let’s examine how fulfillment robots have evolved, exploring the waves of progress in each decade.
We’ll see how these systems have kept improving to handle the massive demands of modern shopping, and we’ll look at predictions of what to expect in the upcoming waves of the artificial general intelligence (AGI) technological revolution.
2000s: Shelf-to-person fulfillment automation dawns
As more people wanted stuff delivered fast, regular warehouses had to keep up. Back then, the main warehouse throughput limitation was that human pickers spent most of their time walking down aisles rather than retrieving items.
In 2003, North Reading, Mass.-based Kiva Systems came up with an amazing idea: using mobile robots to bring shelves to pickers. The company started with Staples as its first customer and quickly added Walmart to the list. The rest is history.
I will not delve into this story since it’s already well-documented. However, to give you an understanding of Kiva Systems’ influence on robotic evolution, it is worth notion that its founders, Raffaello D’Andrea, Mick Mountz, and Peter Wurman, were inducted into the National Inventors Hall of Fame in 2022. This recognition is well-deserved!
2010s: More companies ride the person-to-bin, bin-to-person wave
The next decade started with Amazon.com’s acquisition of Kiva Systems for $775 million in 2012, which led to the creation of Amazon Robotics in 2015. At that time, Amazon stopped supporting Kiva’s client base, leaving a notable technology gap in the market.
Some companies focused on filling that gap with autonomous mobile robots (AMRs), such as GreyOrange, which was founded in 2011, and 6 River Systems, which was founded 2015, acquired by Shopify in 2019 for $450 million, and resold to Ocado in 2023 for $12.7 million. Others saw an opportunity for next-generation solutions.
One alternative approach is person-to-bin, more commonly known as person-to-goods, where fulfillment robots move bins around the warehouse while human pickers fill them in designated aisles. Companies like Locus Robotics (2014) and Fetch Robotics (2014) embraced this strategy.
It’s interesting to note that Locus Robotics spun out of Quiet Logistics, which used to be Kiva’s client and had to find a new way after Amazon stopped servicing external companies. In 2021, Zebra acquired Fetch Robotics for $290 million.
Another approach is bin-to-person, which I divide into multi-bin and 3D versions. Multi-bin representatives include Geek+ (2015) and Hai Robotics (2016), which implemented the idea of a robot with a shelf for multiple bins.
The 3D approach spawned companies like Exotec (2015) and Fabric (2015). The 3D method’s key benefit is higher storage density, which was unavailable by design for shelf-to-person systems without an expensive retrofitting facility for a mezzanine.
Extreme space-efficiency needs led to the development of cube-based automated storage and retrieval systems (ASRS). This niche is mostly led by established automation players, such as AutoStore (1995) and Ocado Group (2000), but also includes modern companies like Attabotics (2015) and Urbx (2016).
From customers’ perspectives, deployment cost and time create the biggest difference between the person-to-bin and bin-to-person approaches. The first method does not require major warehouse modifications and can be implemented without a halt in operations, making it affordable for small and midsize enterprises (SMEs).
The second type, especially the 3D version, heavily relies on custom warehouse infrastructure. As a consequence, it is available mostly to corporations ready for large capital expenditures (CapEx) and to wait for longer return on investment (ROI) periods.
2020s: End-to-end, humanoids, and lights-out waves
Goods-to-person (G2P) systems continue to evolve, with end-to-end approaches such as those of BrightPick (part of Photoneo, 2013), the grocery-focused Fulfil (2017), and my own startup Staxel (2023). Members of this group are trying to solve end-to-end task of order fulfillment by consolidating the order and improving robot unit efficiency.
The “person” part in this “bin-to-person” method differs from the 2010s wave. The consolidated order might be brought directly to a target customer or unloaded to the packaging area for future human processing.
End-to-end approaches promise a vast space for applications besides traditional warehouses. Think of fully autonomous supermarkets, pharmacies, or convenience stores.
And of course, in 2024, it is hard not to mention multi-purpose humanoid robotics startups such as Figure (2022; has raised $754 million to date), Agility Robotics (2015; raised $178 million), 1X (2014; raised $136M), Sanctuary AI (2018; raised $90 million), and Apptronik Systems (2015; raised $29M).
Agility Robotics is currently the most warehouse-focused on this list, given its early pilots with Amazon and GXO Logistics. Of those five, 1X is developing both wheel-based and legged versions, while others solely focused on implementing human legs to achieve promising flexibility in human environments.
It is also worth noting that established companies like Tesla and Boston Dynamics are also publicly working on humanoid robots.
Here is my admittedly biased opinion on the viability of those approaches: End-to-end is a more practical branch of fulfillment robots than general-purpose automation. This could be an “apple-to-oranges” comparison, since humanoids might not be the best for warehouses but might be suitable for factories.
At the same time, several stealth-mode companies like Collaborative Robotics are working in the end-to-end domain and doubling down on a human-capable automation paradigm rather than a human-looking one.
The AGI moment in warehouse automation is lights-out facilities doing the job without human supervision. While I’ve focused on fulfillment robots in warehouses, many other tasks must be automated for a true lights-out scenario, such as piece picking, packaging, truck loading and unloading, and palletizing and depalletizing. Many of those topics are hot in general 202x automation wave, and each deserves a separate article.
Fulfillment robots landscape continues to change
We are moving towards lights-out automation and inevitably have to answer the same question of robots stealing human jobs. From my perspective, human supervision and backup will always be the thing, but they may eventually transform into mainly remote operations. And if you ever launch a robotic project at scale, you will never ask this question again.
For those of you interested in a deep dive into fulfillment robots and warehouse automation for intralogistics, I built a landscape map by segments mentioned above, like shelf-to-person, person-to-goods, bin-to-person, multi/3D/end-to-end, and humanoid robots. I also included other sectors not mentioned in this article, such as nano-fulfillment, package sortation, material handling, shuttles, and industrial-scale solutions.
