At OIA, we take a highly pragmatic approach to AI. Beneath the buzzword lies a host of individual technologies that we’ve been using for some time.
Take reinforcement learning, for example—essentially trial and error—which enables our robotic pick arms to learn to delicately lift items they’ve never handled before.
But of course, AI is still a highly nascent technology. Two years ago, no one had heard of ChatGPT or the benefits of generative AI. Today, we’re actively exploring how these powerful technologies can further help us improve efficiency, lower costs and create a more productive warehouse environment.
How is AI used in Warehouses?
AI in warehouses is already playing a pivotal role in solving some of the most challenging issues, including labor shortages, operational inefficiencies, and increasing demand for speed and accuracy. Here are some of the key areas in which warehouse AI is being adopted:
AI for Inventory Management
One of the most impactful uses of AI in warehouse operations is in managing and optimizing inventory. Traditional, manual methods often lead to overstocking or stockouts, causing inefficiencies and loss of sales. Inventory distortion has long been a key issue.
AI-driven systems can predict demand with high accuracy, using AI models that analyze historical sales data, current trends, and even external factors like weather or economic conditions.
For instance, deep learning could allow warehouse AI to automatically adjust order sizes from a wholesaler to meet predicted demand, minimizing both waste and stock shortages. This level of precision is particularly important in sectors where storage space is expensive, stock is perishable or time-critical, and holding excessive inventory can cut into profits.
Warehouse AI for Order Fulfillment
The rise of e-commerce has placed unprecedented pressure on warehouses to fulfill orders quickly and accurately. AI-powered robotic systems enable automation of monotonous or physically arduous tasks like picking, packing, and sorting, traditionally carried out by human workers.
While warehouse robots have existed without AI for decades, AI enhances these systems by enabling them to make decisions for themselves. For example, a key differentiator between an AGV and an AMR such as Chuck AMR is the latter’s ability to navigate a warehouse floor using computer vision, choosing a path based on data rather than blindly following fixed floor markers. They’re both robots designed to move around a warehouse, yet one has the power of AI to help it optimize routes and think about its work a little more like a human might do.
Warehouse AI robots can also work more safely alongside human workers, speeding up order fulfillment while reducing the likelihood of human error. These systems can adapt to the flow of operations, recalibrating in real-time if a particular robot is delayed or a product is misplaced.
AI for Predictive Maintenance
Downtime is expensive. Predictive maintenance aims to solve this issue by monitoring equipment health in real-time. Instead of relying on scheduled maintenance, algorithms—increasingly using AI—analyze machine data generated by multiple sensors to detect patterns that indicate a potential breakdown.
In many cases, these algorithms can predict which components are likely to fail and initiate preemptive repairs before issues escalate. This helps warehouses avoid costly downtime and ensures that operations run smoothly. Predictive maintenance is not only cost-effective but also improves safety by preventing dangerous malfunctions that could result in damage to stock, infrastructure or even your workforce.
What are the key technologies related to warehouse AI?
Machine Learning (ML)
Machine learning is a subset of AI focused on enabling machines (computers) to learn from data. The quality of the data the machine learns on defines how useful the output is. The most simplistic example is thus: Show the machine 10,000 photos of dogs, and tell it “these are dogs”. Then show it 10,000 photos of cats, and tell it “these are cats”. Then, show it a new photo of either a cat or a dog and it will have learned enough about each animal’s features to give you the right answer. This is also known as supervised learning.
In warehousing, those cats and dogs might be SKUs—and rather than learning their visual attributes, the ML might learn how often they are picked. So when you ask it to prioritize your entire inventory by SKU velocity, it can come back with a neatly organized list.
Deep Learning (DL)
Deep learning is similar to ML, yet far more complex. Using a neural network, it’s able to model complex patterns in data. Returning to our previous analogy, we might show the machine 20,000 photos of dogs and cats jumbled together. Some photos might have both dogs and cats in. Using deep learning, the machine can identify individual aspects of each image, compare them with one another and begin to spot patterns. So when we finally show it an image and say “this is a dog”, it will come back with every image that fits those patterns.
In warehousing, a robot might use deep learning to navigate through its surroundings—learning to recognize and avoid objects that might cause an obstruction.
Generative AI
Generative AI (or Gen AI) is the latest and undoubtedly the most exciting type of AI today. Generative AI is able to generate new content, like text or images, by learning from data analyzed through deep learning. Generative AI powers systems like ChatGPT, enabling them to generate human-like text, create artwork, or even write code by predicting and constructing outputs based on learned patterns.
The applications of Generative AI are still being understood in industry, but the potential is huge. For example, it could be used to design a new warehouse layout, maximizing space utilization and improving efficiency by simulating different configurations and outcomes.
At OIA, we’re actively exploring the use of generative AI to further enhance its operations, with applications ranging from automating customer support to creating entirely new products and services.
Reinforcement learning
Reinforcement learning is essentially trial and error—like learning to ride a bike. Without prior knowledge or data, the machine attempts an action (in the case of our on-grid robotic pick arms, they’ll attempt to pick up an item they’ve never handled before). Invariably, the first attempt will fail. So the machine will attempt to do it slightly differently. And it will keep learning from its mistakes until it has identified the most successful course of action.
Computer Vision and LiDAR
Computer vision is a field of AI focused on giving machines the ability to see. It generally combines one or more vision sensors (aka cameras) with deep learning algorithms designed to enable it to recognize the world around it, or to analyze existing video and images.
In warehouse AI, computer vision could be used to identify damaged goods, count inventory, or ensure that the correct products are picked and packed.
LiDAR is similar to computer vision, using a laser-based sensor to enable spatial awareness.
OIA’s Chuck AMR uses computer vision to navigate warehouses, learn and optimize new routes. If an unexpected object is placed in Chuck’s way, computer vision enables it to identify alternative routes around the obstruction.
OIA’s AI Journey
Ocado Group has been at the forefront of AI adoption in the fulfillment industry for over two decades. As the first fully-online grocery retailer in the UK, Ocado quickly transformed itself into a technology leader, specializing in warehouse AI and robotics.
Ocado realized early on that it needed to develop its own proprietary technology to meet the unique challenges of the sector. Large shopping baskets, perishable goods, and tight delivery windows made traditional retail technology insufficient for grocery e-commerce. This challenge became a proving ground for Ocado’s innovation.
As a result, Ocado’s tech capabilities evolved rapidly, enabling it to develop the Ocado Smart Platform (OSP)—an end-to-end logistics solution that combines AI, automation, and robotics. Initially deployed to support Ocado’s own operations, the platform is now available to other industries, thanks to the launch of Ocado Intelligent Automation (OIA). Today, our expertise extends far beyond grocery, applying our AI-driven solutions to a wide range of industries.
Ocado’s Warehouse AI Solutions
The Grid: Intelligent Storage and Retrieval
At the heart of the Ocado Storage and Retrieval System (OSRS) is the Grid. This consists of a massive grid where thousands of robots move along tracks to pick and retrieve items stored in totes.
The bots are not autonomous; rather, they are guided by the Warehouse Execution System (WES)—essentially our "air traffic control" system. The WES continuously updates the robots on which items to retrieve, where to place them, and how to move around the grid. This orchestration happens in real-time, with the system making split-second decisions to ensure maximum efficiency.
Robotic Pick
Robotic pick is an automated picking solution available 24/7 to alleviate the effect of labor fluctuations and unlock significant cost savings. Combining cutting-edge computer vision and advanced sensing, robotic pick identifies, picks and packs items from storage bins without prior knowledge of what they contain.
Without AI, robotic pick would still function—but we’d need to have carefully programmed the dimension, weight, malleability and grasp point for every item stored in the grid (50,000+ SKUs in our grocery warehouses).
But reinforcement learning gives robotic pick the ability to learn. Using computer vision, the arm will identify what it thinks the object is and give it its best shot at picking it up. If that fails, it’ll try again. And it will keep trying until it has learned the optimal grasp point for that SKU.
This makes the system highly versatile, especially in environments where product ranges are constantly changing. And it’s also a great example of why our cloud-first strategy benefits all our customers; the learnings from each robotic pick arm can be shared across our global user base, so what a robotic pick arm learns in London might affect the pick speed of an arm in Tokyo.
Chuck AMR
Chuck is our collaborative autonomous mobile robot (AMR). It’s a warehouse automation solution that can be used to improve cutaway, picking, sorting and returns tasks.
Chuck leverages AI and ML to prioritize work and optimize picking routes in real-time. Using AI-powered computer vision, it recognizes and learns from its environment, and is able to navigate and optimize its movements.
Responsible AI and Ethical Considerations
As warehouse AI becomes more integrated into operations, the need for ethical guidelines and governance becomes critical. Ocado has developed a robust framework for responsible AI & robotics that ensures that the systems it develops and deploys are transparent, fair, and safe. In a warehouse context this means focussing on transparency and governance.
Ocado’s AI systems are trained on high quality and proprietary data, robustly tested and regularly monitored to ensure they meet strict performance and safety standards. The company also maintains transparency by providing clear and understandable documentation, ensuring that customers and partners understand how our technology works.
A New Era in Warehousing, Powered by AI
The future of warehousing will undeniably be heavily shaped by AI. Across the industry, businesses are embracing AI to streamline operations, reduce costs, and meet the ever-growing demands of modern consumers.
At OIA, we’re embracing AI in all its forms, with the pragmatism required to ensure each warehouse AI application results in a tangible uptick in productivity, transforming our warehouses into intelligent, automated hubs that people like to work in.