ACE Robotics takes aim at embodied AI's data problem

ACE Robotics is using wearable devices to capture how people perform their work, then feeding that footage to the artificial intelligence (AI) software it sells to power robots and other physical AI machines. Founded by Chinese AI company SenseTime co-founder and executive director Wang Xiaogang , who chairs the firm, the company said it is data, not hardware, that differentiates today’s machines from truly useful ones. “The binding constraint for the humanoid and embodied AI industry today is data scale,” the company told Computer Weekly. “After years of effort, the industry has accumulated only around 100,000 hours of manipulation data, far short of what is needed for intelligence to emerge at scale.” Much of that data was derived from teleoperation, where a person guides a robot through a task and the system records the motion. This is extremely inefficient, the company said, because it ties up costly hardware and trained operators for every hour of footage captured. To address the challenge, ACE has developed Ambient Capture Engine 2.0, a system that records people working in kitchens, warehouses and other locations, using a wearable hardware kit comprising a camera-equipped headset, tactile gloves and a motion capture suit , along with annotation support for complex and long-horizon tasks. The offering, which was announced at the World Artificial Intelligence Conference in Shanghai in July, alleviates the need to use robots in the data collection process. “A thousand people wearing our devices can generate 10,000 hours of real-world data in a single day,” the company said. “That’s how we intend to close the gap between the 100,000 hours we have today to the tens of millions the field actually needs.” ACE Robotics also used the conference to show off Kairos 3.1, the latest version of its world model, a type of AI model that gives a robot an internal picture of its surroundings so it can plan its actions. Unlike most large AI models, Kairos runs on the robot, not in a datacentre. The on-device model has four billion parameters, small by current standards, allowing a robot to sense and make decisions without the need to send data to remote servers. “The robot does not depend on connectivity to perceive, decide or act safely,” the company said. The cloud is still used for software updates, fleet coordination and rare cases, it added. Kairos runs on Nvidia chips and has been adapted to other processors, so enterprises are not locked into a single compute vendor. The company has put its technology through its paces in a handful of environments. A system called Xiaoman picks, packs and restocks orders at Alibaba's Taobao Instant Commerce warehouse in Hangzhou. Another sorts and folds hotel laundry. Whether such systems can operate reliably beyond controlled sites remains to be seen, and ACE Robotics does not claim otherwise. For mainstream operational scenarios, it noted, the gap between simulation and reality has narrowed substantially. What remains is “the extreme long tail of open, unstructured environments”. Then, there’s also the issue of cost, seen as the hardest unsolved problem in embodied AI today. “For embodied AI to reach genuine commercial scale, the combined cost of hardware, compute, and deployment operations must fall below the threshold the industry can absorb,” the company said. “It is as much an industrial problem as a technical one.” Omdia, a technology research firm, expects the embodied AI market in Asia-Pacific to hit $1.5bn in 2026, but its chief analyst Lian Jye Su said the technology is still in the hype phase , with large-scale deployments largely limited to consumer applications rather enterprises. Read more about AI in APAC Ant Group has rolled out an AI-powered platform to support doctors at every stage of clinical practice , from online consultations and clinical decision-making to medical research. Singapore’s Personal Data Protection Commission has issued advisory guidelines clarifying when organisations can scrape and reuse personal data to build GenAI models. Less than a year after releasing its national AI framework, Tokyo is updating its guidelines to address the weaponisation of frontier AI models capable of finding and exploiting unknown vulnerabilities. Agoda has set its sights on becoming an AI-powered travel companion as it changes how it builds software and moves its tech workforce into a new facility in Bangkok.