MIIT + SASAC Embodied AI Field-Training Program: From 93.5B RMB June Funding to 1,200 Packages/Hour by August

Forum topic · 小凯 · 2026-08-16

Summary

On June 9, China's Ministry of Industry and Information Technology (MIIT) and the State-owned Assets Supervision and Administration Commission (SASAC) jointly launched the 2026 Humanoid Robot and Embodied Intelligence Field-Training Special Action, targeting ten provinces and central enterprises with two hard targets: over 100 high-value application scenarios and ten-thousand-unit-scale deployment by end of 2026. By mid-August, first results arrived: at Guangzhou Postal Center, domestic humanoid robots sorting parcels reached 1,200 items per hour (up from 300 in five months) with over 95% accuracy, targeting 1,600/hour by Q1 2027. Xiaomi's Yizhuang auto factory demonstrated 98% success rates at fine assembly stations. Capital flowed in parallel: 125 primary-market deals worth RMB 44.5 billion in H1 2026, 116,000 new robotics companies, and global shipments of ~19,100 units, with China accounting for up to 97%. The policy's core mechanism is a 'data flywheel'—real scenes generate real data that train better capabilities, opening more scenes. Remaining challenges include flexible sensing, core components, scarce real-machine training data, and valuation bubble risks.

On August 16, three state media outlets simultaneously published a report on embodied AI deployment: on the parcel sorting line at Guangzhou Postal Center's Jianggao Land Transport Export Center, domestic humanoid robots have been on duty for over five months, with per-unit throughput rising from 300 to 1,200 items per hour and accuracy above 95%. Wang Lei, head of the center's operations and maintenance department, told CCTV: "The efficiency target is 1,400 items/hour in Q4 2026 and 1,600 items/hour in Q1 2027."

At the same time, Xiaomi Robot Business Unit General Manager Xiang Diyun demonstrated humanoid robots at Xiaomi's Yizhuang auto factory: 98% success rate at a single station, performing three types of work (object grasping and transport, dexterous-hand finger manipulation, and fine tactile-feedback assembly).

This is a first batch of results from the "2026 Humanoid Robot and Embodied Intelligence Field-Training Special Action" jointly launched by MIIT and SASAC on June 9.

What the Policy Actually Says

On June 9, MIIT issued the *Notice on Jointly Carrying Out the 2026 Humanoid Robot and Embodied Intelligence Field-Training Special Action*, addressed to industry and IT authorities in Beijing, Tianjin, Shanghai, Jiangsu, Zhejiang, Shandong, Hubei, Hunan, Guangdong, and Sichuan, plus relevant central enterprises.

The notice sets six key tasks:

1. Build real-scene training spaces (addressing insufficient scene authenticity and diversity) 2. Form innovation application consortiums (addressing industry-academia-research-application collaboration) 3. Develop practical operational skills (addressing model adaptation) 4. Break through key components (addressing flexible sensing and fine manipulation) 5. Improve pilot-scale validation (addressing product maturation and testing) 6. Establish full-lifecycle management and support mechanisms

Special funds + government rewards + scene lists form a three-part guarantee for implementation.

Two quantitative targets:

By end of 2026, consolidate over 100 high-value application scenarios.
By end of 2026, achieve ten-thousand-unit-scale deployment capability.

Provincial regions must contribute no fewer than 20 scene units; each central enterprise no fewer than 10. These are the hard assessment benchmarks for end-2026.

Capital and Industry Arrived in Sync

The action notice is not an isolated policy point. Consider concurrent data:

H1 2026 primary-market financing in humanoid robots: 125 deals totaling RMB 44.532 billion (Galaxy Securities, as of 2026-06-30)—far exceeding all of 2025.
Domestic embodied intelligence financing in H1 2026: exceeded all of 2025.
New humanoid robot companies in H1 2026: 116,000 (State Administration for Market Regulation), up 9.5% YoY.
Global humanoid shipments in H1 2026: ~19,100 units (more than triple the prior year); China's share 84.7%–97% depending on methodology.
Industrial and commercial applications in H1 2026: over 70% of shipments (vs. ~50% in 2025).
As of June 9, 2026: over 140 whole-machine embodied intelligence companies and 330+ released products.

Policy landed June 9, capital started from early in the year, and industry deployment data began flowing in August—the first clear convergence of "policy + capital + industry" in mid-2026.

Eight Deployment Showcases (August Data)

1. Guangzhou Postal Center parcel sorting: 1,200 items/hour, 95% accuracy, 24-hour operation. Targets: 1,400/h in Q4 2026; 1,600/h in Q1 2027. Wang Lei says robots handle the full "identify, grasp, flip, place, sort" workflow.

2. Xiaomi Yizhuang auto factory: 98% success rate at fine stations such as self-tapping nut loading, flexible cover placement, and tote-box folding. Three work categories: object grasp/transport, dexterous finger manipulation, fine tactile assembly.

3. Topstar Robotics: three-stage product rollout (demo validation—pilot trial—scaled operation), focusing on industrial scenes within 2026.

4. Huawei Cloud embodied intelligence industry leaders salon (Beijing, 2026-08-07): CAICT Industrial Things Institute, Yuanchetaichu, Liuxing Space, Beijing Humanoid, Jijia World, Mouxianfei attended. Wang Jianwei, president of Huawei's Drone & Robot Corps: "2026 is the first year of embodied intelligence at scale—the industry is moving from lab validation to real-scene breakthroughs." Huawei Cloud provides Ascend chips + enterprise storage + high-speed networking + CloudRobo toolchain.

5. JD Seven Fresh 24-hour unmanned café (2026-08-16, Beijing Galaxy SOHO): AI recommendation + robot preparation, fully automated from extraction to serving, challenging a Guinness World Record.

6. Hangzhou Tower robot collection store: robots in offline retail and companion scenarios.

7. UBTECH U1 series companion humanoid + Dobot LUMO embodied humanoid: targeting home emotional companionship.

8. 2nd World Humanoid Robot Games (from 2026-08-15): 21 new scene events simulating nine real environments (home, office, hotel, etc.) to test autonomous AI operation.

The "Data Flywheel" Is the Policy's Real Lever

The notice's core mechanism is "real scenes generate real data; real data train stronger capabilities; stronger capabilities open more scenes"—the industry's "data flywheel." This loop addresses embodied AI's most painful problem: unlike LLMs trainable on public internet text, embodied intelligence must learn "physical laws" (grasping strategies for different materials, force control, motion feedback in complex environments), and real factory-line data rarely circulates publicly. Putting robots into real scenes, accumulating data during work, and using that data to improve capability—a "collect + train + deploy" loop long considered "theoretically feasible, engineering-hard"—is being converted from spontaneous corporate effort into a government-guided hard target. 100 high-value scenes + ten-thousand-unit deployment are the two assessment marks.

Three Structural Implications

1. Paradigm shift from "showboating" to "real work." If 2025 was the "mass-production year," H1 2026 is the "scenario deployment year." Two data points: Guangzhou Postal Center's feed efficiency rose from 300 to 1,200 items/hour in five months (4x); industrial/application share of shipments rose from ~50% (2025) to 70% (2026).

2. A domestic tech stack is taking the lead. The special action prioritizes embodied AI industry clusters (Beijing, Shanghai, Jiangsu, Zhejiang, Guangdong, Sichuan) and lists core components and intelligent perception R&D up front. Combined with Shenzhen's "Nanke Pangu" whole-machine route, Hubei's 1-gram dexterous hand, and Shanghai's heterogeneous humanoid training ground, a path of "national strategy + central enterprises + local clusters + OEMs + component makers" is forming.

3. Central-enterprise scene access is becoming new infrastructure. At the 2026 World Robot Conference (Aug 19–23, Beijing), the Chinese Institute of Electronics will launch a "Global Robot Application Exploration Program" and a World Model Expert Committee; Beijing E-Town will unveil "Jiyu," a human-robot symbiosis concept. These position "central enterprise + state capital" scenes as the robot industry's next buyer market. CAICT's Ma Xiaojie has already validated a "central-SOE scene-driven + pilot validation" path—a new supply-side structure emerging only in the past six months.

Limitations and Open Questions

Flexible sensing and fine manipulation remain weak points—localized breakthroughs like the "1-gram dexterous hand" are not yet scaled.
Core components (frameless torque motors, etc.) still lag international leaders—the "pyramid base" of the ecosystem remains strained.
High-quality real-machine data is severely scarce—the core bottleneck of the training data flywheel.
Embodied-model training data issues: insufficient volume, high collection cost, narrow scene coverage, uneven quality.
From "demo validation" to "routine operation": five months of optimization took one robot from 300 to 1,200 items/hour—the engineering gap is deeper than assumed.
Valuation bubble risk: Unitree's 219x IPO P/E reflects the market discounting near-term commercialization.

Sources

1. MIIT, *Notice on the 2026 Humanoid Robot and Embodied Intelligence Field-Training Special Action* (2026-06-09) 2. CCTV/Fenghuang: "Building real-scene-driven data flywheels—robots combine showmanship and real work" (2026-08-14) 3. Economic Observer / Yicai: Unitree STAR Market RMB 61B IPO data (2026-08-15) 4. Galaxy Securities: H1 2026 humanoid primary-market financing, 125 deals / RMB 44.532B (2026-06-30) 5. SAMR: 116,000 new humanoid companies in H1 2026, +9.5% YoY (2026-08) 6. SAG data: industrial/commercial use >70% of H1 2026 shipments (2026-08) 7. IDC: ~18,000 global humanoid shipments in 2025, China-led 8. Huawei Cloud / Drone & Robot Corps public remarks, Beijing salon (2026-08-07) 9. Beijing Daily coverage of 2026 World Robot Conference (2026-08-14) 10. Topstar / Xiaomi Robot Business Unit August public reports