AI News

Apple research team discloses humanoid robot autonomous control system HALO, validated on Unitree G1

East MoneyUpdated Nov 8 09:22

AI summary

The Apple research team submitted the paper "Perceptive Humanoid Motion from Simulation" to arXiv, disclosing the humanoid robot autonomous control system HALO and validating it on the Unitree Technology G1 humanoid robot, enabling motions learned in simulation to be executed in unfamiliar real-world rooms. In the first half of 2026, financing in China's embodied intelligence sector reached approximately 43.8 billion yuan, of which embodied brain companies absorbed 22.253 billion yuan, accounting for 50.8%. AgiBot GO-3Preview raised the phone packaging success rate from 50.8% to 98.3%, and XPeng IRON completed the second-generation VLA on-device deployment.

Body · AI translation

Apple's research team discloses progress on robot "brain" research. On November 4 local time, Apple's research team submitted the paper "Perceptive Humanoid Motion from Simulation" to the preprint platform arXiv, disclosing the humanoid robot autonomous control system HALO. The research team deployed the system on Unitree Technology's G1 humanoid robot for verification, enabling the robot to carry the motions learned in a simulated environment into an unfamiliar real-world room for execution.

The robot brain is the advanced decision-making hub of an intelligent robot system, responsible for handling complex cognitive tasks and intelligent decision-making. It is composed of various processors, memory, and algorithms, and is mainly used to process and analyze data from sensors and make corresponding decisions based on that data.

A Zhongtai Securities research report points out that the brain is one of the core variables for embodied intelligence to move from demonstrations to practical use. Multiple robot manufacturers have publicly cited insufficient generalization capability as the biggest constraint, and technical routes are also competing around the integration of VLA and world models. Capital, policy, and industrial strategy are converging on the brain within the same time window, reflecting the industry's recognition of the judgment that "the threshold for the body is falling, while the brain determines the ceiling." On the capital side, funds are concentrating toward embodied brains. In the first half of 2026, domestic financing in the embodied intelligence sector was approximately 43.8 billion yuan, of which companies with embodied brains as their core business absorbed 22.253 billion yuan, accounting for 50.8%, while robot body and complete-machine companies raised only 5.6 billion yuan in the same period, accounting for 12.8%.

Overseas, Physical Intelligence has reached a valuation of 20 billion US dollars. Physical Intelligence (abbreviated as PI or π) was founded in March 2024 by a team of professors from UC Berkeley and Stanford University and Google scientists and other research and technical experts. The company is committed to creating software that can add advanced intelligence to various robots and machines, with the ultimate goal of creating an AI model that serves as a general-purpose robot system (a general-purpose robot brain). The flagship model pi-zero launched by Physical Intelligence is precisely an attempt to solve this difficult problem. The model directly binds images, text, and robot actions, and uses data collected by real robots in physical environments for training.

Domestically, AgiBot GO-3Preview has connected the full-chain learning of pre-training, post-training, and post-deployment, raising the success rate of mobile phone packaging from 50.8% to 98.3%, validating PhysicalRSI in a real business scenario for the first time. XPeng IRON has completed the second-generation VLA on-device deployment, with 0.2-second re-inference and prediction of actions within 6 seconds, and 3 self-developed Turing chips providing 2250 TOPS. Tuoyuan Intelligence has open-sourced the 4D world action model E0.5, with real-machine success rates of 85.0%/88.3%, 13.3/26.7 percentage points higher than π0.5.

Leading robot manufacturers have already made the robot brain a research and development priority. Tesla CEO Musk has repeatedly made clear that for Optimus to truly enter home scenarios, it must solve the three core issues of hands, brain, and mass production, and the "AI brain that understands the physical world" has been listed as one of the core challenges. Unitree Technology founder Wang Xingxing, at the 2026 World Robot Conference, defined "insufficient generalization capability" as the biggest bottleneck in global embodied intelligence, and revealed that the company is conducting preliminary research on a physical AI robot self-evolution system, allowing AI large models to autonomously search academic papers, automatically generate control code, and deploy it to real machines after verification in simulation.

According to incomplete statistics from Cailian Press, among A-share listed companies, those mainly involved in the robot brain field include:

Zhiwei Intelligent: The Zhiqing series Thor robot brain controller has added leading humanoid robot customers such as AgiBot, becoming one of AgiBot's first batch of "A-chain" suppliers. At the same time, the company has carried out joint development cooperation with UBTECH. This technical solution helps reduce complete-machine costs and shrink controller volume.

Tianqi Shares: It holds a stake in Shanghai Junaio Panshi Technology Co., Ltd., which focuses on the research and development of brain-inspired cognitive world models and embodied intelligent robot brains, focuses on the integration route of brain science and artificial intelligence, and builds highly reliable embodied brain capabilities for real scenarios.

Tianzhun Technology: In 2025, the company's output grew significantly, among which the robot brain domain controller and commercial vehicle domain controller businesses in the embodied intelligence solution field increased relatively quickly, and PCB-related LDI products in the visual process equipment field and intelligent equipment in the new automotive field increased relatively quickly.

Horn Auto Electronics: The domain controller developed by the company in cooperation with NVIDIA is a robot brain and can be applied to multiple fields such as humanoid robots, industrial robots, and logistics robots.

ThunderSoft: The company is mainly engaged in the research and development of intelligent robot brain module (chip + operating system + core algorithm) products, providing computing and processing capabilities for intelligent robots.

Nengke Technology: The company's "Lingzhi" embodied intelligence AI training and inference platform targets embodied agents in industrial scenarios. Relying on technologies such as digital twins, physical simulation, and vertical-domain large-model training and inference, it provides training and inference support for embodied intelligence's perception, decision-making, and execution in the physical world, and develops and supports "brain systems" for it.

Kyland Technology: The company's Intewell operating system relies on a microkernel virtualization architecture to achieve the integration of the embodied robot brain and cerebellum, unifying multi-device and multi-node tasks into the AI system and motion system under the architecture, and achieving low-cost, low-power, and small-volume deployment goals through software-hardware integration technology.

(Source: Cailian Press) Middle of the main text Embedded advertisement

Source

East Money