Otto Group and robotics software company Sereact have announced a two-phase plan to put AI-controlled robots to work on one of retail’s hardest manual jobs: inspecting and sorting returned customer orders. Initially, ten AI-controlled dual-arm systems will be deployed at Otto Group’s logistics centre in Poland, with 15 more arms planned in a second phase based on initial results, taking the programme to 25 robots in total.
Why Returns Are So Hard To Automate
Few processes in e-commerce are as hard to automate as returns. Items come back folded, crumpled and repackaged, and the assortment of a multi-category retailer never stops changing. Traditional automation depends on known products and predictable presentation, and neither exists in a returns centre, where the next parcel could contain almost anything.
Sereact’s answer is Cortex, its Physical AI model, which is built to generalise to objects it has never encountered. Instead of relying on a fixed product catalogue or product-specific teaching, Cortex decides in real time how to grasp each item, assess its condition and route it accordingly. “A robot that only knows yesterday’s catalogue is of little use in a returns centre,” said Dr. Ralf Gulde, Sereact’s chief executive and co-founder. “Cortex has to deal with whatever comes out of the next parcel. Every return the systems handle makes the model better at the next one.”
That learning loop is central to the project. The partners will introduce the robots in stages, with each stage generating operational feedback they can use to improve the systems and adapt them to large-scale returns processing. The shared development process is intended to establish whether the technology can meet the reliability, flexibility and quality standards required for broader use across Otto Group’s logistics network.
Trained With NVIDIA’s Physical AI Stack
Cortex is built on NVIDIA’s physical AI stack. Sereact uses NVIDIA Isaac Sim and NVIDIA Isaac Lab to turn real picks from the returns line into thousands of simulated variants, and NVIDIA Cosmos world foundation models to re-render them across garment states, packaging, materials and lighting. New policies are post-trained on NVIDIA Blackwell GPUs and tested against held-out cases before they reach the robots.
The practical payoff, according to the companies, is speed of improvement. When a robot struggles with an item, the fix can come back as a tested update within days instead of weeks, because the lesson from the returns line is first validated in simulation rather than trialled live on the warehouse floor. That loop mirrors the wider shift in robotics, where simulated experience is increasingly used to close the gap between lab demos and messy real-world work, as in LG’s robot data factory with NVIDIA.
People Stay In The Loop
Under a Robotics-as-a-Service model, Sereact will operate and maintain the systems, while Otto Group employees hand over repetitive grasping and lifting and take on supervision, exception handling and quality checks. The companies present that division of labour as a deliberate choice: robots handle the repetitive physical work, while people keep the tasks that depend on experience and judgment.
“Our customers expect returns to be simple and fast – behind that lies a process that is time-critical and laborious, both for customers and for our business,” said Martin Umland, Otto Group’s group vice president for supply chain management. “Returns are notoriously hard to automate, and Physical AI has the potential to change that. This project lets us find out, at real scale, how far Physical AI can take us – an important step towards a more flexible and resilient supply chain.”
A Familiar Partner, A New Test
Sereact is not new to returns work. The Stuttgart company, founded by Gulde and Marc Tuscher, raised a $116 million Series B in 2026, and its customers include Otto Group, Zalando, IKEA, DHL, BMW and Mercedes-Benz. It has previously worked with NVIDIA and Nebius on closing the training loop for robotic returns at CEVA and Zalando, as in Sereact’s earlier robotic returns work.
Why It Matters
This matters because returns are where e-commerce’s promise of easy shopping meets its most expensive reality. Every return has to be received, checked, graded and routed, and doing that by hand at scale is slow and costly, while getting it wrong means reselling damaged goods or writing off stock that could have been recovered. A system that can handle unfamiliar items without retraining could cut that cost and speed up resale, as in South Korea’s push to put AI robots into public-sector work.
It also matters as a test of physical AI itself. Returns centres are unforgiving: lighting changes, packaging varies, and no two parcels are alike. If Cortex can prove itself there, the same approach could spread to other unstructured warehouse jobs. If it cannot, the project will still have shown, at real scale, where the current limits of generalising robots lie. Either way, the next year of data from Poland will be more informative than another polished demo video.
Background
Otto Group and Sereact announced the two-phase returns programme on October 28, 2026, starting with ten dual-arm robots at Otto’s logistics centre in Poland and rising to 25 if initial results support expansion.
Source: GlobeNewswire announcement
0 Comments