Four months old, and it signs Europe's largest supermarket group.

In a 2026 where embodied AI is still stuck between "hot capital, cold deployment," a company called Sota Wujie broke the deadlock with a blitz: a strategic partnership with Europe's largest supermarket group to deploy over a thousand embodied AI robots in real supermarket settings over the next 3 years.

What's more unusual: Sota Wujie doesn't build robot bodies. It builds the "native physical brain" - a 4D world-action model that lets robots truly understand the physical world and act on it.

The founder is Hu Debo, former CEO of Kepler Robotics and a Huawei Europe veteran who fought the European high-end market alongside Yu Chengdong. Chief scientist Liu Zhe, a Huazhong University of Science & Technology PhD and HKU postdoc, has spent a decade on 3D/4D perception and embodied intelligence.

Where Embodied AI Is Stuck: Not "Can't See" but "Doesn't Understand Physics"

Hu Debo's read on the industry is razor-sharp: most embodied systems today are essentially a "migration route."

The mainstream approach trains a vision-language model on massive image-video data, then "grafts" a small amount of robot data to adapt it to action. The result: models can see the world and describe it, but never truly "experience" the physical world - they don't understand how much force a touch needs, or whether turning at this angle will hit themselves, and they produce "physical hallucinations."

Sota Wujie's answer is a native 4D world model: from the training base up, it unifies spatial representation, physical constraints, spatiotemporal reasoning, and action output into one model, so robots don't just "see" the world - they "understand" it and predict the consequences of actions.

In Hu Debo's words, the path to a cross-body, cross-scenario general brain is too long, but "sinking a native physical base model into one high-complexity scenario first, nailing generalization within that vertical, then feeding back to all scenarios" is the most likely path to walk today.

Why Supermarkets? The Hardest Exam Room, the Best Springboard

Supermarkets are more complex than factories: SKUs rotate weekly, displays shift constantly, foot traffic is dense, interactions are diverse. But precisely because of that complexity, they're the best "exam room" for tempering a world model - capabilities forged here migrate upward to home service and downward to industrial logistics.

Data is another moat. Through a self-developed multimodal capture system, Sota Wujie starts with a cold-start 100,000 hours of proprietary operational data; after deployment, it streams back over 1 million hours of real operational data each year, spinning up a "deploy → data回流 → model scaling" flywheel.

Entering European supermarkets also means clearing walls of data security, localized operations, and compliance review - exactly the "home turf" Hu Debo built over years in Huawei's European market.

No hardware, just brains. This seemingly "light" choice is actually a heavier bet: in embodied AI's endgame, hardware will commoditize, and what's truly scarce is the intelligence that understands the physical world itself.

See you tomorrow.