Runway released Praxis-1, an open-weight world action model that converts large-scale video pretraining into robot control policies, with Noble Machines among the partners testing it. In the same batch, UBTECH fitted 112 humanoid units into a 65 m² vertical warehouse, a fleet density that reframes storage layout around robot labor rather than human aisles.
The deployment record is shifting from demonstration to routine duty. AGIBOT's A3 Ultra is handling wayfinding, inquiries and inspection at Jinan Metro, placing a humanoid inside continuous public foot traffic. The same vendor co-hosted Vietnam's first embodied AI conference in Hanoi with the National Innovation Center and local partners, drawing more than 550 attendees from government and academia. Metro duty tests reliability where crowds are unstructured, while the Hanoi event shows ecosystem building traveling through local institutions rather than through unit shipments alone.
On the technical layer, control and sensing remain the scarce inputs. An open-weight control backbone from a video-pretraining lab gives robotics teams a pretrained alternative, which matters most to those without large in-house data pipelines. Sharpa's D01 adds full tactile coverage across interaction areas, and its W02 hand mirrors operator motion in real time, a combination that supports safe human contact and the collection of real manipulation data at human scale.
Capital markets are pricing the same crowded field. Huayan Robotics disclosed an H-share buyback of up to RMB 200 million to be executed on the open market, according to its announcement. A buyback at this stage signals management's read on valuation and cash position while embodied AI listings proliferate.
Two points to watch: whether open-weight control models lower entry costs enough to shift vendor rankings, and whether fleet density metrics, as in the UBTECH warehouse, displace unit shipment counts as the standard measure of deployment.