China’s Spirit AI develops robot software for broader industrial use

China's Spirit AI is developing embodied-AI robot software trained on human video and wearable-sensor data, reporting more than 200,000 hours of interaction data collected and a target of over 1 million hours by the end of 2026, alongside $280 million in funding to expand deployment of its general-purpose robot models. The company said its robots achieved an insertion success rate above 99% for battery test-connector operations at CATL's Zhongzhou production base, with cycle times comparable to skilled workers, and that its wearable collection devices cut data acquisition costs by 90% versus traditional remote robot operation. Founder and CEO Han Fengtao said at the Beijing Academy of Artificial Intelligence conference in June that robot capabilities and deployment costs have not yet reached the conditions for widespread commercial adoption, citing large-scale model training and high-quality data collection as priorities.

China’s Spirit AI develops robot software for broader industrial use China’s https://qazinform.com/tag/china t5478 Spirit AI is developing robot software trained on human activity to expand its use in manufacturing, with battery production among its applications, Qazinform News Agency https://qazinform.com/ correspondent reports. The company’s approach centers on embodied artificial intelligence, which enables machines to interpret their surroundings and perform physical tasks. Its training material includes human video and information gathered through wearable sensors. In a February announcement, Spirit AI said it had collected more than 200,000 hours of interaction data and aimed to exceed 1 million hours by the end of 2026. The company favors varied, unstructured activity to help robots respond to situations beyond carefully prepared demonstrations. “Dirty data is the key to scaling VLA models,” co-founder and chief scientist Gao Yang said in the announcement, referring to vision-language-action models that connect visual information and language instructions with physical movements. Spirit AI also announced $280 million in funding to expand the deployment of its general-purpose robot models. It said its wearable collection devices had cut data acquisition costs by 90% compared with traditional remote operation of robots. One industrial application involves battery testing at CATL’s Zhongzhou production base. According to Spirit AI, its robots insert and remove test connectors, adjusting their movements and the force applied when handling flexible wiring. The company reported an insertion success rate above 99% for that operation, with cycle times comparable to those of skilled workers. That figure concerns a specific manufacturing task rather than general-purpose robot performance. Broader commercialization still faces constraints. At the Beijing Academy of Artificial Intelligence https://qazinform.com/tag/artificial-intelligence conference in June, Spirit AI founder and CEO Han Fengtao said robot capabilities and deployment costs had not yet reached the conditions needed for widespread commercial adoption. He identified large-scale model training and the collection of high-quality data as priorities. At the same event, the company demonstrated its Moz1 robot using the Spirit v1.6 model to organize a desk. According to the company, the robot divided a single instruction into several actions and revised its plan when people moved objects or closed a drawer. Gao told the conference that increasing the amount of initial training data could reduce the additional training needed for new tasks. “Over the next 12 to 24 months, embodied foundation models will make tremendous progress,” he said. Earlier, Qazinform reported https://qazinform.com/news/king-charles-calls-for-international-cooperation-on-ai-safety-be0832 that Britain’s King Charles III called for international cooperation on artificial intelligence safety during a meeting with technology executives in Scotland.