The article analyzes the different paths taken by large-model companies such as OpenAI, Google, Alibaba, and DeepSeek in entering embodied AI, arguing that their primary driver is their own growth and narrative needs, the second is defensive positioning, and that data and gateway are long-term goals not yet realized.
Orbbec's 2026 Q3 report shows robot vision solution revenue reached more than double the year-earlier level, cumulative robot customers exceeded 1,000, and the delivery cycle was compressed from about 40 days to 20 days. First-three-quarter revenue was 711 million yuan, down 0.43% year on year; net profit attributable to shareholders was 73.8615 million yuan, down 31.62% year on year; gross margin rose from 42.8% to 49.09%; and R&D investment was 175 million yuan, up 19.20% year on year.
At its 2026 Innovation Day, AgiBot's Peng Zhihui proposed the robot energy-efficiency unit Worken, defined as the number of active robots × autonomous working hours × output efficiency ratio, to measure how much work robots actually do. At the Longqi factory, 8 robots worked a cumulative 528 hours with an output efficiency ratio of 0.7, for a total of 2,956.8 Worken; at Guangzhou Metro, 120 robots worked 240 hours with an efficiency ratio of about 0.5, for a total of 14,400 Worken.
Robots from about 90 teams competed in the first World Mountain Robot Challenge on Mount Tai, on a 3.9-kilometer course with over 600 meters of elevation gain, requiring them to race or carry fragile items such as eggs and watermelons to the summit. A quadruped robot from Beijing Lingyi Technology (a Unitree subsidiary) won the quadruped speed category in 43 minutes 41 seconds, while a humanoid robot from the Beijing Zhishen Lingyue team completed a shorter course in 27 minutes 26 seconds. The event was hosted by China Media Group and included awards such as Fastest Robot and Best Load-Bearing Quadruped Robot.
The 2026 Physical AI and World Models Conference was held in Nanjing, where Andrew Yao noted that reliability is the threshold physical AI must cross to move from demonstrations to real-world deployment. In September, Kuangda Technology released its first-generation flexible fabric skin system solution for embodied intelligent robots, which has already been delivered in small batches to some leading companies; the second generation is advancing multimodal sensor integration and operating-condition validation. GGII (Gaogong Industry Research Institute) forecasts that by 2030, global demand for flexible tactile sensors for humanoid robots will reach 1.525 million square meters, with a market size of about 27.4 billion yuan.
Primary-market investors are collectively pausing embodied AI projects. IT Juzi data shows 33 financing deals worth 10.105 billion yuan in October 2026, down more than two-thirds from the June peak of 55 deals and 30.892 billion yuan. Secondary-share transfer quotes generally range from 65% to 80% of the previous round's valuation, and 8 of the 13 robotics-related companies that listed in Hong Kong in 2026 broke issue price. The article argues that capital is not exiting the direction of physical-world intelligence, but rather the valuations and delivery pace that demo capabilities cannot support.
After embodied AI funding reached 322 deals worth nearly 93.5 billion yuan in the first half of 2025, September funding fell 40% from August, and the first half of October totaled only 4.9 billion yuan. The article analyzes the cooling from three angles: technology, orders, and valuation. In the first half, 218 publicly disclosed winning bids totaled about 1.7 billion yuan, of which industrial orders accounted for only 20% and were mostly prototype testing. After Unitree's listing, its share price fell from 1,100 yuan to 404 yuan, and more than 20 robotics companies collapsed within 19 months. The author argues that capital has not left but has shifted toward early-stage rounds and up and down the industry chain, with more emphasis on real delivery and orders.
The research treats complete closed-loop robot policy implementations as reusable execution experience: a coding agent generates policy code from a few successful demonstrations and iteratively improves it using simulation feedback. Across four source tasks in RoboCasa, iterative optimization raised the success rate from 28.3% with initially generated code to 64.2% with validation-selected code, and the learning approach was validated on a physical Unitree G1.
ROOF proposes a reinforcement learning method with contact-supported geometric escape guidance that enables proprioception-only humanoid robots to first create clearance and then complete the rise under overhead obstacles. In simulation, the recovery success rate was 88.5% under plates that can only be lifted, compared with 18.9% without geometric guidance; it was 98.6% under free plates with randomized size, mass, and placement. After deployment on a Unitree G1, the policy successfully recovered in 95 of 150 attempts across seven real-world obstacle scenes, including furniture and multi-object arrangements not seen in training.
The authors propose REA, a routed embodied agent runtime that lets a reasoning-focused System 2, an embodied router System 1, and reactive System 0 policies cooperate within a single runtime, with the core rule that a running policy never waits on the brain.
The authors propose ∇Mimic, which uses exact gradients from a differentiable simulator to train whole-body motion-tracking policies for humanoid robots, reaching a score of 0.82 in 20 minutes and 0.874 in 1.8 hours on a single GPU, whereas the world-model method takes 1.2 hours to reach 0.82.
Why it matters: The paper presents a complete scheme for training humanoid policies with exact gradients under differentiable simulation and open-sources the engine and code, making it easy to compare the efficiency of first-order and zeroth-order methods.
DiffWBC proposes replacing deterministic policies with a one-step diffusion model for general humanoid motion tracking, requiring only one forward inference per 50 Hz control step, achieving a local tracking error of 20.1 mm on the BONES benchmark, and deployed on a physical Unitree G1.
Embodied AI companies are investing heavily in the robotics supply chain as industrial capital. Unitree Robotics and Sequoia China established the "Shanyu Innovation Program," and Zhiyuan Robotics appeared in 47 investment and financing events from August 2024 to October 2026, most of them directed upstream of its own complete machines. The article argues that three overlapping uncertainties—unconverged technology routes, an unformed supply chain, and unvalidated scenarios—are leading top companies to use capital to lock in supply chains, technology options, and industry positions in advance, as the industry moves from free competition toward an oligopoly rehearsal.
Robotics concept stocks swung lower, with Unitree Robotics falling over 10%, Daye Co., Ltd., Zhengyu Industrial, and Tianqi Co., Ltd. hitting the limit down, and Shandong Mining Machinery and ASD falling over 7%.
Desktop robots are becoming a new offshoot for embodied AI makers, focused on two functions: emotional companionship and productivity tools. The Letianpai desktop robot entered the market at a price of 1,799 yuan, the Lingtong Nian NIA-F01 offers companion interaction in a figurine form, and Elephant Robotics' myCobot series of small robotic arms requires secondary development. The article argues that consumer products still center on emotional companionship and have not yet truly delivered on boosting productivity, and that the future requires moving toward modularity and upgrading the ability to operate in the physical world.
Unitree Technology's Q3 2026 report shows revenue of 1.817 billion yuan for the first three quarters, up 55.71% year on year, and net profit attributable to shareholders of 387 million yuan, up 270.54% year on year; it is the company's first financial report since going public. In the same period, AgiBot released the Genie G2 Max, a safety-certified heavy-load embodied industrial robot, with a 20 kg payload per arm, 40 kg with both arms, and a 50 kg peak, aimed at depalletizing and palletizing, material transfer, and production-line feeding. Tsinghua University, Infinigence AI, and AgiBot jointly proposed a new paradigm called Collective Experience Learning (CEL) and completed large-scale validation on a cluster of 96 real robots.
Based on a survey of more than 20 mainstream quadruped robot projects on GitHub, the article maps the open-source quadruped technology landscape across five schools: kinematic control, full-stack engineering, optimization algorithms, reinforcement learning, and hardware platforms.
Agility Robotics will list on Nasdaq on November 5 via SPAC under the ticker AGLT, with a pre-money valuation of $2.5 billion and confirmed 2025 revenue of only $1.782 million, about 63.5% of which comes from related parties.
Why it matters: Using Agility's listing filing data as a sample, it breaks down humanoid robot unit economics and RaaS contract structures, and provides transferable metric definitions.
Shandong-based flexible fabric tactile sensing company Juqiao Industries announced on October 29 that it completed a 100-million-yuan Series A funding round, led by Ant Group, with Lingyi Venture Capital, Weihai Gangrong, and Qingdao Hexin participating. The article states that it weaves functional fibers into sensing networks, and that in September 2025 its 1,000th set of humanoid robot fabric electronic skin rolled off the production line, with customers including Unitree, Songyan Power, and Lingcifang; however, its single-point accuracy is inferior to vision-based tactile solutions, making it better suited to large-area coverage scenarios such as the torso and soles of the feet.
Unitree Technology founder Wang Xingxing has appeared on the 2026 Hurun Rich List for the first time with a fortune of 69.5 billion yuan, ranking 78th. Unitree disclosed its first financial report after going public on the evening of October 28: revenue for the first three quarters was 1.817 billion yuan, up 55.71% year on year, and net profit attributable to shareholders was 387 million yuan, up 270.54% year on year. In the third quarter, however, revenue was 665 million yuan, up 69.93% year on year, while net profit attributable to shareholders was 113 million yuan, down 17.37% year on year; R&D investment was 127 million yuan, up 249.58% year on year.
SenseTime's Astra reaches 100% on LIBERO-Long++ and 81.9% on RoboDojo; the stronger the general-purpose brain, the more easily a robot body that conforms to standard interfaces can be swapped out as a "peripheral." Using Unitree's real-machine data flywheel, Zhiyuan's 3/4 R&D investment in the large and small brains, and Galaxy General's AstraBrain-WBC 0.5 trained on 2 billion frames of motion-capture data (92.58% zero-shot motion-tracking success rate, 0.39 ms inference latency) as examples, the article proposes three criteria: depth of in-house development at the execution layer, data return rate, and scenario replacement cost.
Shanghai has proposed deploying 8,000 humanoid robots into factories by 2028, while Jiangsu, Shenzhen and other regions are also advancing real-scenario training and production-line deployment; companies including SAIC and CATL have disclosed application progress. Most deployed cases are concentrated in decomposable processes such as cell grabbing and loading, test loading and unloading, and material handling and sorting. For example, AgiBot's "Nengzai No. 1" works at about one unit every 2 seconds on Buick's battery mass-production line, and Galbot S1 has entered CATL's production line to handle material handling and sorting. The article notes that entering factories still requires passing hurdles in stability, battery life and cost accounting, and cross-factory replication also requires reducing technology adaptation and system integration costs.
Hangcha Group's third-quarter operating revenue was 5.024 billion yuan, up 7.45% year on year, and net profit attributable to shareholders was 575 million yuan, down 9.76% year on year. Multiple scenarios at its Hengfan plant have completed trial runs of unmanned forklifts, and forklift robots are expected to be brought to market next year. Lanjian Intelligent signed a warehousing and logistics equipment project for phase two of Wuhan Chuyang New Energy, valued at about 258 million yuan. Inge Intelligent launched the Messol-B embodied intelligent loading and unloading robot, equipped with its self-developed OmniPredict 2.0 loading and unloading world model.
KAIST and other institutions are developing biomimetic robotic animals such as robot beetles and robot ants to solve practical problems including rescue and detection. A cyborg cockroach at Nanyang Technological University in Singapore carries a microcomputer, gas sensors, and an infrared thermal camera, and is remote-controlled via electrode pulses; a single roach can survey about 100 square meters of rubble, but a full survey of a building would require about 500 roaches and 500 operators. A robot ant at Stanford University can carry a payload 2,000 times its own mass, grips using a gecko-inspired foot design, and requires wired power in industrial settings due to its high energy consumption.
Using Unitree H1, Isaac Sim 2024.1, IsaacLab, and PPO as its stack, the article walks through a complete sim-to-real pipeline: converting URDF to USD, scene setup, domain randomization configuration, PPO training, ONNX export, and ROS 2 deployment on the real robot.
The team recreated a real self-service tea station as a physically interactive 3D environment in Isaac Sim, placed a Unitree G1 in it, and used a script to execute the five-step tea-making motion. The robot's motions were not extracted from human video; the entire run took place in simulation with physics enabled. The G1 still shows imperfect contact and slipping when interacting with objects, and there is currently no learned policy or automatic trajectory generation. The team is exploring whether such environments can serve as a source of synthetic robot demonstration data, including for robots with different embodiments.
The study proposes REA (Routed Embodied Agent), a unified runtime that lets a reasoning-focused System 2, an embodied routing System 1, and reactive System 0 policies collaborate within the same framework, with the core rule that a running policy does not wait on the brain.
Unitree Technology plans to invest 20 million yuan through its wholly owned subsidiary Hangzhou Unitree Robotics Co., Ltd. in the Hangzhou Embodied Intelligence Application Pilot Base, in which it holds a stake. This round brings a total of 200 million yuan in new capital from five parties. After the increase, the pilot base's registered capital rises from 100 million yuan to 300 million yuan, while Unitree's stake remains at 10%. The funds are mainly for the base's project construction and business development. In the first nine months of 2026, the pilot base reported operating revenue of 449,600 yuan and a net loss of 15.2047 million yuan.
During a live demonstration at the "2026 World Physical AI Conference and Unitree Partner Ecosystem Conference," Unitree Technology's H2-X humanoid robot, just released on October 21, lost balance and fell sideways while attempting a side somersault landing, and one mechanical lower leg detached from its connection point. The robot is officially named "H2-X Kung Fu Legend Guardian" and is positioned for security and emergency rescue scenarios, with a maximum battery life of about 8 hours. Official promotional videos have shown it kicking in car doors, lifting vehicles, and performing a somersault after jumping down from a platform about 4 meters high. As of the time of reporting, Unitree Technology had not commented on the incident.
Robots are already providing services in several consumer scenarios in Beijing, including an AI pancake-making robot at the Longfusi food street, the Xbot coffee robot at Beihai Park, the Galbot vending robot at a Qianmen subway entrance, and the Robot Flame Research Institute 2.0 restaurant in Yizhuang, covering dining, retail, and interactive experiences. In July this year, the Municipal Bureau of Commerce included embodied intelligent robots in the trade-in subsidy program, offering a subsidy of 15% of the final sale price, capped at 1,500 yuan per item; the Yizhuang robot 4S store sells more than 50 products, and the Unitree Wangfujing direct-sales store mainly promotes the Go2 robot dog.
D-Robotics and Unitree Robotics have formed a partnership to jointly build robot solutions for education scenarios, starting with the Sunrise S100 computing platform and later expanding to the full Sunrise S and X platform lines. The Sunrise S100 is a mid-range compute platform for embodied intelligence, offering 80-128 TOPS of on-device inference compute and supporting multi-camera input and local inference of multimodal large models. The first phase will create standardized training solutions for universities, vocational schools, science and innovation competitions, and youth AI education; after the education scenarios are validated and mature, the partners will expand to industrial scenarios.
DeepSeek founder Liang Wenfeng debuted on the 2026 Hurun Rich List at No. 7 with 280 billion yuan in wealth, leading this year's newcomers. He has recently focused intensively on the embodied intelligence field and spoken with multiple entrepreneurs, and is considering whether DeepSeek should move further into this track; options may include forming a team himself or entering through investment. People familiar with the matter said he judges that DeepSeek still lags GPT-6 in some capability dimensions, and that scaling language models alone may not be enough to support AGI competition.
Dingjiao One analysis says Unitree Technology's share price has rebounded by more than 70% cumulatively in recent days, with market capitalization recovering to over 280 billion yuan. However, its first post-IPO financial report, released on October 28, shows Q3 revenue of 665 million yuan, up 69.93% year-on-year, while non-GAAP net profit was 105 million yuan, down 17.42% year-on-year.
D-Robotics and Unitree Technology exhibited together for the first time at the Changsha Education Equipment Exhibition and announced a strategic partnership, launching an embodied intelligence education solution with the Sunrise S100 computing box as the core computing foundation, combined with Unitree's Go2 quadruped robot and R1 humanoid robot. The solution includes 15 themed courses totaling 180 lessons, covering elementary through high school, and offers a two-level progression path (desktop level and real-robot level) plus 6 courses integrating competition and coursework. The Sunrise S100 integrates a CPU, BPU, and MCU within a single SoC, placing perception and decision-making and real-time motion control on the same chip for coordinated operation.
D-Robotics and Unitree Robotics exhibited together for the first time at the Changsha Education Equipment Exhibition and announced a deep collaboration, launching an embodied intelligence education solution built on the Sunrise S100 compute box as the core compute foundation, combined with Unitree's Go2 quadruped robot and R1 humanoid robot. The solution covers hardware devices, a curriculum system, course-competition integrated practice projects, and teaching environment construction, targeting primary and secondary schools for embodied intelligence education. The two parties are taking education as their first implementation scenario, with plans to build a replicable regional robotics education demonstration model and expand into industrial and other scenarios.
Humanoid robots are moving from launch-event stages to real-world settings such as the Canton Fair and cinemas for public testing. Guangzhou Zhidong Future deployed robots across five scenarios at the Canton Fair simultaneously, while Dobot's Atom worked 14 consecutive hours in a single day at a Shenzhen cinema and sold over 1,000 cups of popcorn.
Desktop robots are becoming a new offshoot for embodied AI vendors. Roboparty launched a bipedal desktop robot at 1,799 yuan, Lingtong's NIA-F01 is a humanoid figurine companion robot, and Elephant Robotics' myCobot series takes the form of small robotic arms aimed at secondary development. These products reuse the smartphone and IoT supply chains and cost less than full-size humanoid robots; consumer-grade selling points currently center on emotional companionship, while efficiency-tool solutions mostly require secondary development.
Unitree Technology disclosed its first quarterly report since going public on the evening of October 28. Revenue for the first three quarters of 2026 was 1.817 billion yuan, up 55.7% year on year, already exceeding the 1.699 billion yuan for all of 2025. Gross margin for the first three quarters fell to 54.0%, with the third quarter alone at 50.6%, below the 60.4% for all of 2025; net profit attributable to shareholders after deducting non-recurring items was 349 million yuan, down about 18.8% year on year.
Unitree Technology released its 2026 third-quarter financial report on October 29, with single-quarter revenue of 665 million yuan, up nearly 70% year on year; non-recurring net profit attributable to parent-company shareholders of 105 million yuan, down 17.4% year on year; and an estimated gross margin of about 50.6% for its main business, below the 56% in the first half. In the third quarter the company invested more than 127 million yuan in R&D, up nearly 250% year on year, and disclosed a plan to subscribe 20 million yuan for an equity stake in the Hangzhou Embodied Intelligence Application Pilot Base to maintain a 10% holding; the center has already collected and open-sourced its first batch of real-machine datasets based on Unitree robots.
Unitree Technology's revenue for the first three quarters of 2026 was 1.817 billion yuan, up 55.71% year on year, while non-GAAP net profit was 349 million yuan, down 18.78% year on year; R&D investment was 263 million yuan, up 191.48% year on year. The article places this report of revenue growth without profit growth alongside JPMorgan's October 25 upgrade to Overweight with a target price raised from 300 yuan to 995 yuan, and Wang Xingxing's statement that the window for a physical AI breakthrough may come in the next one to two years or two to three years, noting that valuations are pricing in platform value after 2028, while current revenue still comes mainly from research and education scenarios.