AI Robots Take On Unpredictable Returns
Otto Group's Polish centre will trial up to 25 physical-AI systems that handle unfamiliar items, with expansion depending on early results.
Updated on Nov. 2, 2026 in Robotics
Live Poll
Do you think AI-controlled robots taking over warehouse work is good for workers?
Otto Group is deploying up to 25 AI-controlled robotic systems at a Polish logistics centre, starting with ten dual-arm systems. Partner Sereact will operate and maintain them under a Robotics-as-a-Service agreement; the deployment is announced and its results have not yet been reported.
Why it matters
The deployment targets the difficulty of processing unpredictable e-commerce returns, which vary in condition, packaging and orientation, a problem that has resisted conventional fixed-catalogue automation. It sits within Otto Group's €350 million technology and AI investment commitment and signals a move toward robots that handle unfamiliar items.
The announced figures are 25 maximum systems and 10 in the initial phase, but no performance benchmarks such as throughput, grasp success rate or error rate have been published. The deployment is described as a Robotics-as-a-Service arrangement, and the only quantified investment is the €350 million group-wide commitment.
The players
Otto Group
A German multinational retail and logistics group whose logistics network is the site of this AI robotics deployment.
Sereact
A German robotics company whose Cortex Physical AI model drives the systems, and which will operate them under a Robotics-as-a-Service agreement.
NVIDIA
A chipmaker whose Isaac Sim, Isaac Lab and Cosmos tools generate the simulated training environments and whose Blackwell GPUs support model development.
The details
The system is built on Sereact's Cortex Physical AI model, which identifies, grasps and assesses items it has never seen before, without relying on a predefined product catalogue. Training happens in simulation: NVIDIA Isaac Sim and Isaac Lab, along with the Cosmos tools, generate virtual environments in which the model practises handling objects, and NVIDIA Blackwell GPUs support model development and testing. Human staff shift from repetitive handling toward supervising the robots, managing exceptions and carrying out final quality checks.
Timeline
Article on the Otto Group and Sereact deployment published 2 November 2026.
The Tech Race
The project positions Sereact's physical AI approach against other efforts to move robots beyond fixed-programme tasks into open-ended handling, using NVIDIA's simulation stack as the training backbone. Its reliance on simulated environments mirrors a wider industry pattern of training general-purpose manipulation models before field deployment.
Shoppers returning e-commerce goods may see faster processing at Otto Group's Polish logistics centre once the systems are running, though no timeline for that has been given. The human staff who handle returns will shift toward supervision and quality checks, and the timing of the second phase depends on early results not yet announced.
The takeaway
Watch whether Otto Group confirms the second phase of 15 additional systems after early results are reported. That decision will indicate whether Sereact's physical AI can handle unfamiliar returns reliably enough to scale across the wider logistics network.
Further reading
For background on the technology behind this deployment, explore the latest coverage in Robotics.
Source note: This article includes information reported by Intelligentcio.
Live Poll
Do you think AI-controlled robots taking over warehouse work is good for workers?
