(Yicai) Nov. 13 -- China’s energy sector is emerging as a commercial testing ground for embodied artificial intelligence, with state-owned energy giants such as China National Petroleum Corporation and China Southern Power Grid forming partnerships with the country’s leading robot makers.

CNPC’s industrial investment platform plans to set up a joint venture with two companies -- Unitree Robotics and a startup incubated by the Chinese Academy of Sciences that is working on world models for embodied AI -- to develop embodied AI solutions for industry, Beijing’s market regulator announced recently.

Unitree’s initial public offering documents, published in August, outlined potential collaborations with CNPC, including its quadruped robots being deployed across CNPC’s oilfields, refineries, pipelines, storage tank farms, and other places to carry out safety inspections, detect leaks, and perform environmental assessments following incidents.

The “physical AI brain” of embodied intelligence systems can analyze aggregated data centrally to quickly identify equipment faults and anomalies, and transmit monitoring results back to a control center in real time, making inspections much more efficient than if they were done manually, Pang Haitian, founder of Jiangxing Intelligence, told Yicai.

Robot dogs, drones, and inspectors can completely replace human personnel at remote power facilities, where inspections are labor-intensive, Pang noted.

China Petroleum and Chemical, better known as Sinopec, has also begun using robots at plants, deploying a smart inspection unit at the 12,000-square-meter liquefied gas spherical tank farm of its Zhenhai refinery last year. The robot can collect 15 types of data, autonomously inspect the area 12 times a day, and identify potential equipment leaks within a 50-meter range, the firm said.

The nation’s power sector is also becoming a key industrial area for the use of embodied AI robots. China Southern Power Grid and Unitree teamed up in September to push forward their use in power-related operations. And in a recent competition, the grid operator required robots to complete tasks such as inspecting transformer substation equipment and cutting electricity supplies during emergencies in a real-world environments.

Embodied AI for industry is transitioning from prototype demonstrations to engineering delivery, while some high-value applications have successfully completed pilot testing, though mass-market deployment remains some way off, according to Guo Tao, an angel investor and senior AI expert.

Technological breakthroughs and a “commercial shakeout” will happen simultaneously in the sector, Guo said. He expects a major wave of iteration across key areas, including the physical generalization capabilities of AI models, the reliability of hardware platforms, and the overall cost of the systems.

In addition, capital is expected to gravitate toward companies with proven and stable delivery capabilities along with the ability to replicate solutions across different operational environments, he said. Meanwhile, many players that can only complete prototype production but cannot secure sustained and long-term orders from factories will be whittled out of the market, Guo pointed out.

Editors: Tang Shihua, Martin Kadiev