Ant Group's Lingbo tests one robot brain on a real pharmacy aisle

Ant Group's robotics unit Lingbo used the 2026 Bund Conference in Shanghai to stake what it calls the "brain over body" thesis on a live Shanghai Guoda pharmacy aisle. The story is not the keynote. It is the test the vendor just wrote itself into.

The named test is a real one. Lingbo says it has built a "smart pharmacy" inside Shanghai Guoda Pharmacy retail locations, and that the deployment is live in stores with no fixture retrofit. The work happens in aisles roughly 80 centimeters wide, where boxes of similar size stack in dense rows, packaging is often transparent or reflective, and individual SKUs land on shelves in arbitrary order. A robot looking for one specific box has to pick the right one out of a row of lookalikes under those lighting and layout conditions. That is the friction the model has to absorb if the cross-embodiment claim is going to mean anything outside a staged demo, per the Leiphone (雷锋网) writeup of the conference.

The model at the center of the pitch is LingBot-VLA 2.0, branded by Lingbo as a "native embodied general brain." It follows the Vision-Language-Action (VLA) pattern common to current robotics research, where a single network reads camera and language input and outputs joint commands. The vendor says the pre-training pass has already absorbed data from 17 mainstream robot brands and 20 body configurations, spanning single-arm, dual-arm, bipedal, and wheeled robots. The showcase pairs that single model with two physically different bodies on the conference floor: a Leju (乐聚) humanoid for the logistics scene, and a Titanium Tiger (钛虎) unit for an industrial production line. Same model, different hardware, the same Leiphone writeup reports.

That is the part the wire will carry. Three things are not in the report, and they are what a reader should hold against the claim.

First, the 17-brand and 20-configuration numbers come from Lingbo itself. The Leiphone coverage is a single-source event recap, and no competitor or independent lab has yet scored the model outside the showcase floor. Until a third party runs the same body through the same aisle, "general brain" is a vendor-attested generality, not a measured one.

Second, the Shanghai Guoda partnership is named, but the deployment is described from the vendor side. The writeup does not include a Guoda spokesperson quote, throughput figures, error rates, or a count of which stores are actually running Lingbo's stack today versus staged for the conference. "No fixture retrofit" is the strongest on-record claim, and even that lives only in the vendor's framing.

Third, the wider competitive frame is being imported from the conference program. Lingbo's chief scientist Shen Yujun (沈宇军) is scheduled to take the main-stage "Physical Intelligence: Disagreements and Choices" roundtable, and the side forum on "embodied ecosystem co-building" lines up with the vendor's positioning. The challenge series being launched jointly with ModelScope and Alibaba Cloud Tianchi is, by definition, a forward-looking announcement, not a result.

The Bund Conference is the venue, not the verdict. The reason the venue matters is that a real pharmacy deployment, however thinly sourced, is the rare case where a general-embodiment claim has to clear a non-curated physical test rather than a lab benchmark. Aisle width, packaging variance, and SKU disorder are the kind of friction that breaks a model trained on tightly staged data. If Lingbo's stack keeps working through a working week at Guoda, the cross-embodiment framing starts to look like a product, not a pitch.

The next visible checkpoint is the Ant Lingbo Embodied Large Model Challenge, which the company is announcing jointly with ModelScope and Alibaba Cloud Tianchi. The submission rules and held-out test set, when published, will be the first independent scoreboard the vendor has agreed to play on. Until then, the score is the pharmacy aisle, and the aisle is the only one a reader can hold against the headline.