On September 3, at the Yibin International Convention and Exhibition Center, under the spotlight of the World Power Battery Conference, 55 robotics teams completed an unusual "interview": the applicants were robots, the interviewers were production lines, and the questions were real tasks spanning production-line handling, goods delivery, clothes folding, beverage preparation, emergency rescue, and more. A total of more than 130 teams registered, covering four major tracks: industrial, commercial, lifestyle and entertainment, and open innovation.
This is the "Jiangyuan Cup" 2026 Embodied Intelligence Scenario Skills Challenge. What makes it especially notable is that teams brought working robots, not PowerPoint slides; industry partners asked not about creative concepts but about pricing, delivery timelines, and customization options. The competition thus became a job fair, the arena became a proof-of-concept (POC) validation site, and trophies could turn into purchase orders.
The tasks came directly from the production lines of local Yibin companies. The industrial track included production-line handling and feeding, parts sorting and assembly, and industrial campus inspection, with fixture locating pins required to stay within a tolerance of 0.1 mm—roughly the thickness of a human hair. The commercial track simulated campus delivery and exhibition-hall guiding, while the lifestyle and entertainment track focused on flexible manipulation, precision grasping, and autonomous decision-making. The open innovation track was entirely open, including firefighting robots and other applications.
Applause on the exhibition stand cannot bridge the gap on the production line. Changing lighting on the line can cause vision systems to fail, mechanical vibration can make inertial measurement units drift, and cramped workspaces can set multi-axis coordination at odds with itself. The industry generally faces a three-layer gap: insufficiently concrete scenarios, data hunger, and a lack of validation. Universities have algorithms but lack scenarios; companies have scenarios but lack technology; and academia and industry lack a mechanism for dialogue under the same set of rules.
Yibin's solution is to open up industrial scenarios across the city and create a closed loop through data feedback. The Southwest Embodied Intelligence Training Center officially began operations in September 2025. Under its plan, by 2026 its professional data-collection operations team will exceed 500 people, deploy more than 250 data-collection robots, and aim to produce more than 200,000 hours of high-quality multimodal interaction datasets. For StarWar Technology, training and evaluating models on such real-world scenario data aligns with the approach behind AI training platforms: letting learners and teams practice on real problems and make engineering trade-offs under computing-power and budget constraints.
Capital and talent are also falling into place. Yibin has partnered with multiple institutions to establish eight AI-focused funds with total assets under management of RMB 3 billion, and has already invested in more than 20 robotics-related companies. The city is home to 14 universities and 100,000 college students, supplying talent to the industry. Under its plan, by 2028 Yibin will cultivate three to five data-service companies with 1,000 employees each and more than five chain-leading embodied intelligence companies, driving an output value of RMB 10 billion to RMB 15 billion.
Perhaps the best way to get robots out of the lab is not to give them a bigger exhibition booth, but a real job. The competition will end, but their "employment" has only just begun. Whether scenarios, data, capital, and talent can continue to converge will determine whether this city's model can be replicated.