Robots Pack Groceries: Inside Ocado's Automated Warehouse

The first thing that strikes a visitor inside Ocado’s sprawling customer fulfillment center (CFC) in Luton, just outside London, is the silence. Unlike the typical industrial din, there’s no clatter of machinery, rumble of engines, or chatter of human voices. Instead, a low, pervasive hum of electronics fills the air, a fitting backdrop for a facility where robots significantly outnumber people.

Ocado, an online grocery company launched in 2000, has always championed automation as a means to navigate the notoriously tight margins of the grocery business, especially for online orders which incur additional picking, packing, and shipping costs. Though you might not be directly familiar with Ocado, its technology powers online grocery operations for major retailers like Kroger across 14 US states, Sobeys in Canada, and both Morrisons and its own delivery brand in the UK, alongside other clients in Europe and Asia.

At the core of this automated ecosystem is “the Grid,” a vast, crisscrossing network of tracks occupying most of the warehouse’s top floor. Here, a fleet of hundreds of blocky robots glides seamlessly, orchestrated by a central computer to prevent collisions. These robots efficiently transport custom-built trays laden with groceries—from bread and tins to ready meals—to their designated locations. The level of automation is so complete that from a maintenance walkway, human presence is almost entirely absent, save for the Ocado employees guiding visitors. Even technical support for the robots is handled remotely by a team based in Bulgaria; a robot signaling a problem with an amber light will often self-correct and resume operation within seconds, requiring no direct human intervention.

While the Grid system itself isn’t new, having been a cutting-edge marvel when first observed in 2018, it is now undergoing a significant upgrade: the integration of robotic arms. Previously, the cuboid bots on the Grid primarily moved crates, delivering items to human workers who then manually packed them into shopping bags. This system, while efficient, still relied on human dexterity, with workers expected to pack items in mere seconds, guided by a display optimized to minimize unnecessary movement.

The new addition, dubbed On-Grid Robotic Pick (OGRP) arms, marks a significant leap. These stationary robots are strategically positioned across the Grid, each equipped with a small suction cup. At the Luton facility, 65 OGRP arms work in concert with 500 of the original Grid robots, which bring both customer shopping bags and incoming groceries to them. The arms then pick up individual items and meticulously pack them. Each OGRP arm incorporates a camera for picking, though they currently lack the capacity to identify damaged goods like broken eggs or bruised apples, a task still best handled by humans.

In 2024 alone, fewer than 100 OGRP arms packed over 30 million orders. Ocado anticipates deploying nearly 500 such arms by the end of this year. James Matthews, Ocado’s deputy CEO, notes that these arms currently handle approximately 40 percent of Ocado’s groceries. The company aims to reach an 80 percent automation rate within the next two to three years, a goal facilitated by the development of new end-point attachments beyond the current suction cup, including parallel grippers and soft, hand-like devices.

Complete automation, however, isn’t the objective. Ocado acknowledges that some items, like heavy wine bottles or watermelons, are too cumbersome or fragile for current robotic capabilities, and developing dedicated tools for every single item isn’t economically viable. Yet, the learning capabilities of these systems are remarkable. Years ago, the challenge of packing a bag of oranges seemed insurmountable due to their unpredictable movement and the risk of damage. Today, the underlying AI models—which Matthews describes as “cousins” to the generative AI models making headlines elsewhere—have autonomously learned to grasp the label of an orange bag, lifting the entire package without incident, a skill now replicated across the robot fleet.

Ocado is continuously exploring automation opportunities throughout the warehouse. While humans still unpack incoming shipments and load products onto the Grid, and others load heavy trolleys onto outgoing vans, the company is actively developing new automated solutions for these tasks, including mobile robots. The one area less likely to see full automation anytime soon, according to Matthews, is last-mile delivery. Despite investments in autonomous driving startups, the customer-facing nature of deliveries makes full automation less appealing, as it would likely shift the burden of unloading onto the customer.

Ocado’s business model is particularly well-suited to automation because its relentless pursuit of efficiency has already streamlined many jobs into simple, repetitive tasks, making them ideal candidates for robotic takeover. Moreover, some of the roles being automated, such as packing frozen foods in sub-zero temperatures, are notoriously difficult to staff. Matthews candidly states, “You just literally cannot find the people who want to come in and work in a freezer,” making these positions prime for automation.

Crucially, Ocado maintains a degree of separation from the direct employment implications of its technology. It sells its CFC systems to clients like Kroger, who then employ the human workforce. When layoffs occur, they are administered by the client, not Ocado, which itself continues to grow, opening new sites, expanding R&D, and hiring more remote support staff.

The future for grocery stores supported by Ocado clearly involves further automation and fewer human jobs. Beyond the current robotic arms, the next wave of innovation focuses on “efficiency” itself. Ocado is developing lighter, cheaper, and more energy-efficient robots, including new 3D-printed Cartesian models that weigh a third of their predecessors. Lighter robots not only consume less energy but also reduce the risk of damage in collisions, allowing for more compact and modular Grid designs that can scale down to smaller sites.

Matthews believes the most significant future changes will be less predictable, driven by advancements in analytical intelligence rather than physical hardware. The real challenge lies in designing machines intelligent enough to navigate “edge cases”—the 10 percent of problems, like a bent crate causing a jam, that currently require expensive human intervention. “It isn’t useful solving something 90 percent of the time,” Matthews contends, “Because if 10 percent of the time you have to pay an expensive engineer to go and unjam it, you’re better off doing it manually.” These remaining “10 percent problems” represent the final frontier for robotic ingenuity, a gap the machines are rapidly closing.