Most industrial policy sets a direction and leaves the specification to the market. Beijing’s new humanoid robot plan sets the specification, down to the price of a hand.
On 11 November the Beijing Municipal Bureau of Economy and Information Technology issued the Beijing Embodied Intelligent Robot Industry High-Quality Development Action Plan for 2026 to 2028. Several clauses read less like policy than like a sourcing document.
The targets
By 2028 Beijing intends to cultivate at least five internationally first-class and ten domestically leading humanoid robot manufacturers, with embodied intelligence models, bodies and key technologies at an international standard.
Capacity is more concrete. The Beijing-Tianjin-Hebei region is to hold 25 per cent of national assembly capacity, key component localisation is to exceed 60 per cent, and Beijing alone is to reach 200,000 units a year across three to five internationally competitive platforms.
The plan also positions the city as a technology source rather than an assembly cluster, a phrase that matters because it implies owning the parts other regions cannot substitute.
The numbers carry the policy
The dexterous hand must achieve a rated load life of one million cycles, fingertip force control resolution of 0.1 newtons, and a unit cost within RMB 10,000. A cost ceiling written into an industrial policy is unusual in China.
Tactile sensors must resolve 0.01 newtons and one millimetre spatially. Six-axis force sensors must hold full-scale linearity error at or below 0.1 per cent with dynamic response above 5kHz. A bionic face must respond within 50 milliseconds and reproduce more than 200 expressions.
Whole machines face their own thresholds: payload at 40 per cent of their own weight, mean time between failures above 2,000 hours, annual failure rate of core motion parts below 10 per cent, design life beyond five years, and more than four hours of continuous operation. Batteries must reach 350 watt-hours per kilogram with 1,500 cycles.
Borrowing the car supply chain
Liang Hongjun, deputy director of the municipal bureau’s intelligent manufacturing and equipment division, framed the plan around four transitions: laboratory to production line, assembly to system-level design optimisation, product development to volume deployment, and technology to enterprise building.
The supporting measures follow that list. On data, more than one million hours of high-quality data across at least 100 scenarios, feeding a national data infrastructure. On compute, more than 50,000 petaflops of dedicated capacity and pairing arrangements with computing cities outside Beijing.
On testing, the plan proposes the world’s first test ground built on real scenarios and multi-body adaptation, run on a principle where user organisations set the questions and testing bodies answer them.
The manufacturing chapter is the most revealing. Beijing wants at least ten shared machining platforms across the Beijing-Tianjin-Hebei region covering structural parts, surface treatment, SMT placement and wiring harnesses, cutting sample delivery to one or two days. That is the maturity of the automotive supply chain, imported wholesale into robotics.
Scenarios are quantified too: at least 20 industrial, 20 service and 10 special-purpose benchmark deployments, a robot innovation park at the Olympic Green, and more than 200 innovative companies with one-to-one support for listing. The document also states, in writing, its opposition to involution-style competition.
Two years to prove the numbers
Detailed targets are not delivered targets. The RMB 10,000 hand and the 2,000-hour failure interval both depend on a supply chain that is still learning to hold tolerances at volume, and 2028 is two years away.
The shift in emphasis is still real. The question in Chinese embodied intelligence has moved from whether a machine can be built to whether it can be built to a specification. Once the specification exists, the procurement decision becomes comparable across suppliers, and that is where margins get decided.
Editor’s note: This is an adapted translation of the original OFweek report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-11/ART-8321201-8480-30711456.html.