AI "Takes Over" the Digital Trade Expo: Wang Xingxing Discusses Millimeter-Level Errors in Robots, Anheng Information Demonstrates "Using AI to Counter AI"

① Wang Xingxing believes that whoever can solve this millimeter-level error problem of alignment with the physical world in the future will have basically solved the core challenge of robotics. ② In the era of large-scale agent applications, truly AI-native autonomous confrontation determines the future of cybersecurity. 《STAR Market Daily》 September 26 (Reporter Xu Cihao) From September 23 to 27, the 5th Global Digital Trade Expo was held in Hangzhou.

This year's Digital Trade Expo took "Meeting the AI Future at the Digital Trade Expo" as its annual theme, with an exhibition area of 170,000 square meters and more than 2,000 participating companies, including 5 STAR Market-listed companies. In the exhibition halls almost entirely "booked out" by artificial intelligence, STAR Market companies took the stage at several of the most talked-about booths, showcasing cutting-edge innovations such as embodied intelligence, AI security, high-end computing power, and industrial intelligence.

A mecha priced at more than 3 million yuan: Wang Xingxing responds to why it "must be built"

In the theme pavilion of the Digital Trade Expo, the booth with the longest line was Unitree Technology's GD01 manned mecha—the world's first mass-produced manned mecha, which not long ago appeared on the cover of the U.S. magazine Time together with Wang Xingxing.

This mecha is about 3 meters tall, painted in red metal, with a semi-open cockpit on top, and weighs about 500 kilograms when carrying a person. According to on-site staff, the GD01 relies on Unitree's self-developed high-precision servo drive system and transformation control algorithm to switch stably between different forms, providing new solutions for scenarios such as cultural tourism performances and special operations.

"Do you want to buy it? The price is more than 3 million yuan." When asked about the price by a 《STAR Market Daily》 reporter, on-site staff said that many people had asked about the price, but not a single unit had been sold yet.

At the Digital Trade Expo's Digital Trade and Artificial Intelligence Dialogue Forum, Unitree Technology founder Wang Xingxing delivered a keynote speech titled "From Machinery to Intelligence—The Evolution Theory of the Embodied Future."

In his speech, Wang Xingxing explained why this three-meter-tall manned mecha was developed. "As the underlying technologies of AI and robotics mature, the R&D and deployment of large robots and small robots do not conflict. Large robots are an unstoppable trend in the industry. Even if Unitree does not do it, other companies will launch similar products in the coming years."

"We define the GD01 as the off-road vehicle among robots," Wang Xingxing said in the speech, noting that it is not for indoor urban use but for complex outdoor terrain and strong passability, to complete tasks in field environments.

In Wang Xingxing's view, the entire embodied intelligence and humanoid robot technology has only developed for three or four years and is still in the early stage of the industry. It will still take several years before robots enter households on a large scale and become widely used by civilians.

It is worth noting that Wang Xingxing judges that the embodied intelligence track will inevitably usher in its own "ChatGPT moment" in the future. When robots can complete 80% of tasks in 80% of unfamiliar scenarios through voice and embodied capabilities, it will mean the industry has reached the critical point of explosive growth. At that time, global companies and resources from various countries will pour into the track, and this industrial inflection point is expected to arrive in the coming years.

"But the biggest bottleneck in the industry right now is the insufficient precise matching between the input and output of AI models and the real physical world," Wang Xingxing said in the speech. Large language models run in digital space, where text input and output have very small deviations and almost no physical loss. Robots are different: perception, decision-making, and motor execution all produce deviations at every step, and the deviations continue to accumulate.

Wang Xingxing said that when robots perform large movements such as carrying and moving, the overall direction is generally fine. But in the final stage of fine operations such as grasping and assembly, errors of several millimeters appear. It looks as if the movement is already in place, but this tiny error cannot be automatically corrected, directly causing task failure and a sharp drop in the overall success rate. "In the future, whoever can solve this millimeter-level error problem of alignment with the physical world will have basically solved the core challenge of robotics."

Anheng Information Chairman Fan Yuan said in an interview with the media during the Digital Trade Expo's West Lake Forum that AI security is a systemic project, not limited to the model layer, but covering the entire chain system of agents, MCP, Skills, and endpoint devices. Only by achieving unified and systematic management can a secure foundation be built for the era of AI autonomy.

Fan Yuan believes that AI autonomous technology is reshaping the cybersecurity industry landscape. "Whether it is Jensen Huang's judgment based on computing power use that AI will naturally enter the field of software vulnerability discovery and cybersecurity, or the several AI autonomous attack incidents that have occurred internationally one after another, they all point to one trend: in the era of large-scale agent applications, truly AI-native autonomous confrontation determines the future of cybersecurity."

At this expo, Anheng Information showcased at the Digital Trade Expo China's first AI-native code audit agent—the Hengnao AI Code Audit Agent—as well as the world's first AI-native honeypot product capable of capturing autonomous AI Agent attacks—the Lieying·AI Threat Hunting Decoy System—turning "using AI to counter AI" from a concept into a deliverable product.

In addition to the cutting-edge robotics and AI security tracks, many STAR Market science and technology innovation companies presented their core technologies. Joulwatt, a STAR Market-listed company in the field of server CPU power supply, brought AI server power supply demos such as VRM multiphase power supplies, E-fuses, and POL point-of-load power supplies, covering key links from power supply architecture to load points in AI computing power equipment.

Supcon Technology, through display screens at its booth, focused on the intelligent development of the process industry and the construction of autonomous operation factories, showcasing its self-developed time-series large model TPT—taking AI from "being able to chat" to "being able to run a factory."

In addition, Moore Threads showcased its "cloud-edge-device" full-scenario intelligent computing product matrix during the Digital Trade Expo, including the MTT S5000 AI training and inference integrated intelligent computing card and the MTT C256 super node, among which the MTT C256 super node uses a single layer of Scale-up network to turn 256 GPUs into a data-center-level computer.