Understand this story
On November 5, 2026, Wang Xingxing, founder of Unitree Robotics, said at the Hongqiao sub-forum of the China International Import Expo that the core challenge currently facing physical AI is the alignment between models and the physical world. He pointed out that humanoid robots already possess relatively strong understanding and planning capabilities, and that what is really holding back large-scale deployment is the "last few millimeters" of alignment error between multimodal models and the input and output of the physical world. Wang Xingxing expects that this link is likely to be well resolved within the next one to two years. A November 6 report by Shangguan News again cited his judgment, saying that the biggest current challenge for embodied intelligence is likewise that the input and output between multimodal models and the physical world are not well aligned, with a relatively large deviation remaining in the last few millimeters. At present, this judgment remains a statement made on the same occasion, and no subsequent progress has been disclosed.
AI-generated from public reports
Developments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 7
Shanghai Observer
CIIE's "intelligence" content hits a new high: AI shifts from exhibit to hardware foundation and factory rhythm
At this year's CIIE, AI is no longer displayed separately; instead, it has become the hardware foundation, the factory rhythm, and the robots' "brain," with the narrative shifting from "how big the model is" to "can it get work done." Siemens' industrial AI Agent Eigen has improved execution efficiency by 2 to 5 times compared with manual workflows and has been deployed at hundreds of Chinese manufacturing enterprises; AgiBot's Expedition A3 humanoid robot achieved 106 consecutive rallies in a back-and-forth table tennis match with a human. AI is also serving the exhibition itself in reverse: robot tea-brewing, photo-taking, and exoskeleton rentals have become part of the daily scene on site.
Nov 6
East Money
A-share Hotspot Outlook: Wang Xingxing Discusses Physical AI Alignment Challenges, Panasonic Copper Clad Laminate Factory Explosion, XFusion to Appear at ChiNext Listing Review on November 12
Wang Xingxing, founder of Unitree Robotics, said at the Hongqiao International Economic Forum that the core challenge of physical AI is the alignment between models and the physical world, and that the input-output error problem may be resolved in the next one to two years. Panasonic's Hsinchu Hukou copper clad laminate plant in Taiwan, China, exploded and caught fire on November 4; the plant's MEGTRON series capacity accounts for about 20% of the global total, and institutions estimate it may affect 4% of global high-end copper clad laminate supply. XFusion's ChiNext IPO plans to raise about 8 billion yuan and will go before the listing review committee on November 12; its 2025 revenue was 58.246 billion yuan and net profit 1.030 billion yuan.
Shanghai Observer
Wang Xingxing: Robot Combat Is Less Entertaining, and the Biggest Challenge for Embodied Intelligence Is Aligning Models with the Physical World
Wang Xingxing, founder of Unitree Robotics, said that the biggest current challenge for embodied intelligence is that the inputs and outputs of multimodal models are not well aligned with the physical world, with significant deviations remaining in the final few millimeters. He pointed out that improved robot stability has actually made robot combat less entertaining to watch than last year, and mentioned that GPT-6, with zero training, directly controlled a Unitree robot to draw. He expects this alignment problem to be largely solved within the next one to two years.
Nov 5
Shanghai Observer
Unitree Technology's Wang Xingxing: The Biggest Bottleneck for Humanoid Robots Is Alignment with the Physical World, Not Model Understanding
At the Hongqiao sub-forum of the China International Import Expo, Unitree Technology's Wang Xingxing judged that humanoid robots already possess strong understanding and planning capabilities; what is truly holding back large-scale deployment is the "last few millimeters" alignment error between multimodal models and the physical world's inputs and outputs, and he expects this link to be solved within the next 1–2 years. In the first half of this year, China's humanoid robot shipments already exceeded 40,000 units, accounting for 97% globally, and are expected to surpass 100,000 units for the full year. Unitree's G1 has become the world's best-selling humanoid robot; in 2024 it was already deployed in automobile factories, with a success rate approaching 100% for fixed-station processes.
Yicai
Wang Xingxing Predicts Five Core Resources of the Future; Turing Award Winner Diffie Says AI Will Sweep the Globe
Wang Xingxing, founder of Unitree Robotics, proposed at the 9th Hongqiao International Economic Forum that data, computing power, chip production capacity, energy, and minerals will become the five most core resources of the future world, and that competition around computing power will be "very, very fierce." He believes that multimodal models are not yet fully aligned with the input and output of the physical world, and that precision operations still have errors of a few millimeters; this bottleneck is expected to be resolved within one to two years. Whitfield Diffie, winner of the 2015 Turing Award, said that the relationship between humans and robots will become the most important issue of the 21st century.
YicaiRelated discussion
Yicai joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Phoenix Technology
Wang Xingxing on the Biggest Bottleneck of Physical AI: Alignment Between Models and the Physical World, Possibly Solvable in the Next 1 to 2 Years
Wang Xingxing, founder of Unitree Robotics, said at the 9th Hongqiao International Economic Forum that the core challenge of physical AI today is the alignment between models and the physical world, and that related problems are expected to be solved in the next 1 to 2 years. He pointed out that multimodal large models already possess strong understanding, planning, and navigation capabilities, but when running on real robots, sensor measurements at the input end are inconsistent with the true state, and action commands at the output end do not match actual motion, with errors being particularly pronounced in contact-based tasks such as grasping and assembly. IDC data shows that in the first half of 2026, global humanoid robot shipments approached 25,000 units, a year-on-year increase of 432.1%, with shipments in the Chinese market exceeding 19,000 units, accounting for about 77.9% of the global total.
Heat trend for this story
- Current
- 6
- Period peak
- 28 Nov 5 23:00
- 24-hour change
- -65%
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.