The Otto Group, the largest online retailer of European origin, is deploying the world’s most innovative automation strategies to improve the performance and accuracy of its supply chain and deliver an even more reliable experience to its customers.
When people talk about the incredible potential of AI, a majority of the conversation has been focused on its language capabilities. But could something like ChatGPT be applied to robotics?
The answer is yes. ChatGPT is an AI application built on a foundation model. Covariant has taken a similar approach to building a foundation model with reinforcement learning to revolutionize the handling of objects in the physical world.
To build a foundation for future growth, the Otto Group is entering into a long-term strategic relationship with Covariant. In the Otto Group’s fulfillment centers, more than a hundred AI robots will be deployed to handle the order-picking process. The start is planned in the centers in Haldensleben and Altenkunstadt.
The Otto Group automation journey: Started with the "feet"; now the "hands"
We are in the midst of unprecedented retail transformation. As limitations in the labor market continue to increase while the scale of e-commerce volume also continues to grow, the need for automation has become even more paramount.
To increase efficiency, build better resilience to labor market challenges, and improve the quality of jobs that exist within their facilities, Otto Group has focused in the past on adopting warehouse robotics to automate the “feet”.
Yet the handling of items — picking, placing, and sorting — what we call the “hand” part of the work — has largely remained manual. Robotic arms that could perform picking and placing activities have existed in principle for decades. But they have failed to perform autonomously (until now) due to the complexity and variability of the items flowing through any given fulfillment or distribution center.
The challenge for a picking robot is that it must manipulate a dynamically changing variety of goods in shape, color, and quantity without master data. Autonomous robotic picking can only be successful at scale when powered by the right type of Artificial Intelligence. Ultimately, you need an AI that enables the robot to see, think, and act based on the unique situation it’s facing.
With the Covariant partnership, the Otto Group will leverage the first foundation model for robotics to achieve autonomous robotic picking at scale. Raphael Adrian Maier, Group Vice President of Supply Chain Management, explains, “AI Robotics offers us the possibility of responding to several challenges in the area of logistics. The skills shortage can thus be compensated for. But first and foremost, they enable us to continue operating our fulfillment centers close to our customers. This strengthens Europe and even more so Germany as a location for doing business.”
The lightbulb moment: Robots that learn, together
Otto Group Board Member for Service Kay Schiebur and Raphael Adrian Maier recall that while initial meetings were very promising, the decisive step was a visit to the Covariant lab in Emeryville, CA.
Maier comments: "We watched the robot arms sorting objects. This is a familiar activity for people: We can recognize things and know how to pick them up. A machine has to learn that. The way the robots handled an apple was very impressive. At first, the arm was unable to grip it. It, therefore, rolled the apple around a little to get a feel for it, after which it succeeded in lifting it. On its next attempt, it succeeded immediately. We could see that the AI was undergoing a learning process."
What’s unique to Covariant is that rather than building specialized models that are trained on a small number of items, Covariant builds the Covariant Brain, a single foundation model trained on everything. And that ultimately delivers a level of generalization never seen before in e-commerce fulfillment. The Covariant Brain, Covariant’s universal AI platform, has been trained on millions of objects from warehouses around the world. This enables the robots to handle the widest variety of SKUs and items possible, from small cosmetics to apparel polybags to fresh groceries.
Additionally, all Covariant robots learn together as a fleet since they are connected to the same universal AI brain – ensuring that learnings and operational improvements automatically propagate across the entire Otto Group network.
Today, a Covariant induction robot can perform up to 1600 picks per hour — i.e. inducting items onto sorters, conveyors, and auto-baggers. And this performance level is just at the beginning – we expect even more increases in efficiency in the next few years.
A joint commitment to the future: Investing in performance and innovation
The vision of this partnership is to have hundreds of Covariant’s AI-powered robotic solutions installed across all Otto Group fulfillment centers.
While the impact of these robotic deployments will be seen on Day One, working with a pioneer like Covariant is a long-term investment. Kay Schiebur stresses that “the point is not just to make logistics a little more effective. It is still largely unforeseeable what further possibilities there will be for the use of general AI in or outside of logistics. However, it is clear that we are facing that future with courage and curiosity."
Schiebur emphasizes that the partnership with Covariant is a typical Otto Group investment, as it resonates with two core values of the company: performance and innovation. “When a difficult market environment prevails, it is important that investments in the future do not fall by the wayside. We are investing here and now in our future viability, in order to strengthen our position as the market leader in Europe.”