Sota Wujie's native 4D model, which learns from real shelf footage across three spatial dimensions plus time, is its bet against simulation first rivals, though the European retailer is unnamed.
Sota Wujie, founded in April 2026, says it has signed Europe's largest supermarket group to host training of its embodied-AI robots, a four-dimensional world model the company describes as "native" to real retail rather than simulated first. The retailer's name is not on the record in any Chinese-language coverage, and the figure most often repeated, more than 1,000 robots deployed over three years, is a company-stated plan, not a contracted fleet.
Founder Hu Debo, an ex-Huawei Europe executive and former Kepler Robotics co-founder, has built his third venture around a vertical-first training strategy. Chief scientist Liu Zhe, a HUST PhD and HKU postdoc with more than 2,380 Google Scholar citations, anchors the technical side. The pair frame supermarket aisles as the highest-fidelity bridge between factory robotics and home service, and Europe's data-localization environment as a structural moat around the dataset.
The bet runs against the simulation-first approach most major embodied-AI efforts have converged on, including Wayve, NVIDIA's Cosmos and GR00T, Physical Intelligence, and Tesla. If real-shelf data does not generalize beyond the aisle, the strategy loses; if it does, the company holds a vertical dataset competitors cannot easily replicate.
Sota Wujie signed the project at WAIC 2026 as a Shanghai municipal AI key project.