On September 21, ACE Robotics (大晓机器人) and PetroChina Shanghai Sales Company announced a strategic partnership to place embodied intelligence robots into gas station convenience stores for a "full closed-loop trial operation." This is the first robot project deployed in a gas station convenience store scenario in China, pushing the route from "robots demoing in flagship stores" to "robots actually working in 24-hour real-world locations." The cooperation follows a four-part framework: "scenarios set the problems, platforms coordinate, models drive, and robot bodies execute" for an integrated warehouse-store solution.
A Concrete Scenario: Gas Station Convenience Stores
PetroChina Shanghai Sales Company manages a network of gas station convenience stores covering all of Shanghai — 24-hour operation, random customer flow, dense shelving, and unpredictable order peaks. This environment puts three pressures on robots at once:
- Product recognition: many SKUs with similar packaging.
- Picking and transport: narrow aisles, tall shelves, precise lifting required.
- Order fulfillment: sudden peaks, low fault tolerance, coordination with human cashiers.
- Chairman: Wang Xiaogang (SenseTime co-founder and executive director)
- Chief Scientist: Academician Tao Dacheng (member of the Australian Academy of Science / world-class AI scientist)
- Standardization: store layouts and operational processes are highly consistent, making replication easy.
- Long tail: locations on highways, in suburbs, and remote areas where chain convenience stores are usually unwilling to set up.
- 24-hour operation: night shifts are hard to staff, and robots can work continuously.
- AgiBot Zhengyuan A3 Ultra: multi-robot collaboration in 4S dealerships, hotels, and convenience stores (shelf organizing, inventory counting, heavy-item transport).
- Galbot G1 (Galaxy General): entered a FamilyMart in Beijing's Zhongguancun in April, handling grilled sausages, cold drinks, coffee, and hot food.
- JD.com "Intern" robots: entered JD Mall stores in June for greeting, guiding, and shelving; its JoyAI-RA model achieves 90%+ success on single-item picking.
- Single-store scaling: this trial did not disclose specific store or robot counts. Real proof awaits a review of the first 10+ stores.
- Cross-store generalization: can picking habits Kairos learned in store A transfer directly to store B?
- After-sales and spare parts: human takeover procedures after robot failures, spare-parts networks, and maintenance cycles are the hidden bills of real deployment.
- Regulation and safety: robots working near fueling areas face stricter electromagnetic compatibility and explosion-proof requirements than ordinary convenience stores.
The robot body ACE Robotics provided this time is a self-developed wheeled-arm lift-type robot: a mobile chassis, an operating arm, and lifting height that covers both high and low shelves. The software side runs "Kairos World Model 3.0" (开悟世界模型 3.0), iterated from the company's first open-source and commercialized world model in China. The signal chain reads order and inventory data from the order system, and the model generates execution sequences for recognition, picking, transport, and packing.
ACE Robotics' Background
The company spun off from SenseTime's robotics division and officially began operating as "ACE Robotics" in 2026. Its board configuration:
The company built its own "ACE R&D paradigm," linking three elements into a pipeline: environmental data collection → Kairos World Model 3.0 → generalized embodied modules. It offers targeted responses to the industry's common pain points: data scarcity, poor common sense, difficult generalization, and insufficient versatility.
The Embodied Super-Brain Module A1 is the company's "general brain" package for multiple scenarios, containing three layers — visual perception, task planning, and a skill library — and can be mounted on different robot bodies for security/inspection/service, retail/logistics, and manufacturing/public services.
Why Gas Stations Were Chosen
The partners' announcement cited a figure: roughly 110,000 gas stations plus over 100,000 public charging stations nationwide. Overlaying these two networks forms an extremely dense offline energy-node network:
Night-shift hiring difficulty is a real constraint. PetroChina Shanghai's choice effectively packages "night shift + repetitive labor + standardized workflows" and hands all three to robots, challenging them to prove they can "stay on duty continuously without human takeover."
How This Differs from Previous Embodied Deployments
Since September, several domestic manufacturers have sent robots into retail scenarios:
ACE Robotics' differentiation lies in two points: the scenario extends from "store core areas" to "the back warehouse and fulfillment chain of gas station convenience stores," and the mode shifts from "robots doing demos" to "robots joining closed-loop fulfillment." The former emphasizes interactive display; the latter connects picking, transport, and packing directly into the order system.
Unavoidable "Next Steps"
Sources
1. GeekPark, "ACE Robotics and PetroChina Shanghai Sales Company Reach Strategic Cooperation" (9-21): four-factor cooperation framework, Kairos 3.0 and A1 modules, company strategy. 2. Shanghai Securities News / Sina Tech / Tencent News same-source reports (9-21): deployment milestones, wheeled-arm lift-type robot body, order-inventory-execution chain. 3. Yangtzeer English compilation (9-22): background comparison and global retail deployment trends.