Nine Departments, 300 Hours of Touch Data, Robots at Workstations

China's nine departments jointly issued guidelines calling for wider deployment of AI, big data and embodied intelligence across commerce, the clearest top-level signal this year that procurement incentives may follow. On the same day, Sharpa released SharpaDex v1.0, 300 hours and 28,993 demonstrations with fingertip tactile sensing on all ten fingers in every frame.

The dataset matters more than its headline size. Manipulation research remains constrained by tactile corpora that are orders of magnitude smaller than vision-language data, and full-hand, frame-synchronised contact sensing is rarer still. If the release holds up under third-party use, it lowers the entry cost for contact-rich policy learning and gives benchmark builders a comparable substrate.

Hardware disclosure is moving in parallel. Jilin University's team published the internal biomimetic structure of a hand it claims carries the highest degrees of freedom in the field, while our own report on HokMind sizes demand and pricing for a 32-DOF bionic face. Taken together, the evidence suggests the degree-of-freedom race is entering a phase where mechanical transparency and unit cost, not peak specification, determine which designs reach production.

Control transfer also advanced. RAI Institute's five-motor bicycle robot ran zero-shot on physical hardware at 8 m/s, completed a half-second flip and cleared a one-metre jump, a hard balance problem where simulation-to-reality gaps usually surface first. Single-platform results do not generalise automatically, but they indicate that agile locomotion pipelines are becoming reproducible outside proprietary labs.

Commercial signals point the same direction. At the 2026 China International Industry Fair (CIIF), exhibitors replaced choreographed routines with workstation deployments, and at the Canton Fair (广交会) participants identified local partners, localised service and compliance as prerequisites for export scaling. Capital is reportedly returning to a long-neglected robotics segment. Two observation points follow: whether post-guideline procurement converts into orders within two quarters, and whether tactile datasets of this scale change manipulation benchmark results rather than only their marketing.