Highly skilled industrial "heroes" — China's embodied intelligent special-purpose robots have arrived. They can scale vertical walls with ease, and their humanoid arms are capable of both welding and grinding, mastering a full range of skills. With VR-enabled remote control operating within milliseconds, high-altitude tasks no longer require personnel to be on-site — the true meaning of technology comes to life! #Technology# #Robotics# #EmbodiedIntelligence# #MadeInChina#

Embodied intelligence can’t afford bad data. Perle Labs is building AI data infrastructure designed for this reality: → Human-verified, expert-validated work → On-chain auditability → Sovereign and enterprise-grade design Built to benefit everyone.

🤖 #Hubei#'s first embodied intelligence talent training base has opened at the Hubei Humanoid #Robot# Innovation Center in #OVC#. The province is home to 10+ academicians in embodied intelligence and 140+ humanoid robot firms. #OVCIndustry# #OVCTalent#

From Lab to Life: Hangzhou Base Accelerates Embodied Intelligence Training A national mid-test base for embodied intelligence in Hangzhou is tackling a key industry bottleneck: a shortage of training scenarios for robots. The 5,000-square-meter facility houses over 140 robots across 40 application scenes, from farming to food service and boxing. Since its launch in May 2026, the base has brought together leading academic societies, top universities, and industry players such as Unitree, Huawei and Alibaba. It offers public tech services covering computing power, data, model validation and scenario testing. “We extend our services throughout China, and even globally, providing comprehensive ecosystem services for the embodied intelligence industry,” says Li Xingteng, whose aim is to turn new tech into good products. #Robotics# #AI# #Hangzhou# #ChinaTech#

🎯 A major milestone on the road to embodied intelligence. This week, @GoogleDeepMind's Gemini Robotics VLA reaches an important milestone in our joint mission to bring AGI into the physical world. By introducing agentic capabilities - the ability to reason, plan, actively use tools, and generalize - we’re moving beyond reactive models and into a new era of robotic autonomy. We’re proud to be collaborating on Gemini Robotics 1.5 with Google as we gear up to deploy Gemini-powered Apollo humanoid robots in additional customer facilities. It’s a critical step in transforming a powerful model into a field-ready system that performs with consistency, reliability, and purpose. Read the blog: #apptronik# #humanoid# #humanoidrobot# #robotics# #google# #deepmind# #geminirobotics#

Excited to share our latest research introducing Qwen-VLA—a unified Vision-Language-Action model for general embodied intelligence 🤖 By combining Qwen3.5-4B with a 1.15B DiT decoder, it unifies manipulation, navigation, and trajectory prediction into a single framework. With embodiment-aware prompts, the same Qwen-VLA model can operate across 11 robot embodiments under a unified architecture—covering single-arm, dual-arm, and humanoid platforms without task-specific policy heads—without task-specific architecture forks or separate policy heads.

A Weifang company is developing educational humanoid robots and AI courses with Shandong University. 🤖 The city's first robot football tournament is being prepared, offering a closer look at embodied intelligence in education.

Just some public market read through from Chinese private VC markets: Institutions are pouring funds into physical AI and world models. 1. Large models / LLMs: ~$23.56B 2. AI infrastructure + technical layer: ~$15.74B 3. Embodied intelligence / physical AI: ~$13.36B 4. AIGC applications: ~$8.79B 5. Autonomous driving + other Top-20 cluster: ~$3.82B, but not apples-to-apples with the above. Some notes: - "Early-stage pure foundation-model funding is basically closed". Looks like more funding is just being put into existing leaders and going into World Model companies. My guess is that we'll likely see the same in the US with Anthorpic/OpenAI consolidation. - "World models have become the biggest consensus in early-stage investment." I said months ago 4D AI/World Models would be the most interesting moving forward, and called out $AEVA as potential exposure. But there's not exactly any pure play exposure. But probably next we'll wait for the next IPOs here in this sector maybe H1 next year. - AIGC application sector is the most mature for AI technology commercialization "Artificial Intelligence Generated Content commercialization is mature but no clear winner yet". Makes sense. in the US there's stuff like Grok Imagine, Google Nano Banana, etc. no clear winner too. especially for video. _ TLDR: Continued funding into AI infrastructure/semiconductor supply chains. Huge capital rotation influx into physical AI / embodied brain / humanoids + world models from capital inflow. Consolidation around leading frontier model companies. Personally just validated what I've been focusing on with Agility Robotics and physical AI players (eg. leaderdrive, harmonic, etc) in public markets... As new potential opportunities in terms of capital rotation. But sadly no world model pure play exposure yet.

Liang Wenfeng’s remarks at the investment meeting: DeepSeek did not start out to make money, go public, or chase the capital markets. It was a group of ordinary people driven by profound goodwill toward the world, wanting to do something useful for humanity. The company is vision-driven, with no formal organizational structure, KPIs, or performance evaluations. The vision is embodied in the way they work and their attitude toward the world; its core is restraint and goodwill. He repeatedly elaborated on “restraint”: the AI market is large enough (potentially 10% of human GDP) that monopoly is impossible—sharing is essential. Open-sourcing is part of both the vision and the business strategy, not something forced upon them. Open-sourcing does not hurt revenue because pricing aims only at a reasonable profit—recovering hardware costs in ten months (roughly a 6x return)—rather than maximizing profit. Demand is inelastic at higher prices; lowering prices made employees cheer because the goal is to make the technology affordable for more people. After open-sourcing, third-party deployment costs remain higher and cannot meaningfully challenge their own business. Last year’s consumer-side users and this year’s enterprise revenue are both by-products on the road to AGI; they have no intention of fighting for traffic or building a super-app. The long-term goal points clearly to AGI. The technical roadmap is clear: after the current Agent stage, the next step is continuous learning (allowing the model to adapt over long periods like a human), then the singularity of self-iteration, and finally embodied intelligence. Continuous learning is the key bottleneck; once solved, AI can accelerate the company’s own R&D, making the path much easier. The company sticks strictly to the main line (language models, CoT, Agents, etc.), declining to pursue video generation, world models, or other off-track directions, while remaining willing to help others. The sole core interest is team stability. Money and resources are not the problem—as long as the people stay, AGI will be achieved. The recent financing has substantially reduced this risk. The main gap with the United States is compute resources (currently about 20,000 H-equivalent cards, with aggressive expansion underway); the talent gap is small. The strategy is to buy as many cards as possible at reasonable prices and prioritize converting cash into GPUs. Domestic chips (Huawei) face a historic opportunity in the ecosystem; technologies such as TileLang can dismantle CUDA’s moat, with validation expected within a year. Organizationally, “official work” occupies no more than half of people’s time; the rest is left for individual exploration. There is little overtime, preserving a relaxed research culture. Commercialization has always been happening, but it is not the goal—focusing on AGI creates a dimensional advantage. Ultimate differences will lie in cost, timing, and user experience; China may carve out a place through lower costs and better experiences. The vision is restraint, sharing, and concentration on the main line, thereby increasing the probability of achieving AGI. Liang Wenfeng carries clear traces of quantitative trading: as a quant expert and the boss of a quant firm, he extracted substantial returns from market volatility and deeply understands that “excess returns are unsustainable” and that “excessive greed will eventually backfire.” This experience has been internalized into a character of “not competing and leaving profits for rivals”—he proactively open-sources top-tier architectures and models, prices only for reasonable cost recovery, explicitly refuses to fight for consumer traffic, avoids head-on competition with big tech, and is even willing to help Alibaba, Moonshot, and other competitors reproduce the technology. This is not feigned humility; it is the use of “restraint” as a strategic weapon: sacrificing short-term sesame seeds to secure the long-term watermelon and team cohesion. He is simultaneously pragmatic and idealistic: on the one hand he candidly admits “we are just a group of ordinary people” with low starting points and scarce resources; on the other he uses vision to unite people and maps a continuous-learning roadmap toward AGI. On the one hand he emphasizes that the company must still survive (enterprise revenue can provide a floor); on the other he treats profit maximization as a dead end that will be defeated by those who take less. His decision-making style is decentralized and consensus-driven, almost obsessive about team stability, yet aggressive in expanding compute—classic quant thinking: control the controllable risks and amplify asymmetric advantages. He can roughly be described as a calm, restrained, long-termist technological idealist: the caution of a market veteran and the pure spirit of a scientist converge in him. He neither fantasizes about monopoly nor shies away from commercial reality, always placing the achievement of AGI above personal and corporate interests. People who truly make money in the stock market eventually become different kinds of good people—donating, sharing, and practicing religion or charity.

Really glad I finally got to visit @XSquareRobot. We got an early look at XRZero-G0, their open-source full-body data collection and training system that works without requiring a physical robot body. At the time it hadn’t been officially released yet, so we couldn’t share anything haha. Now they’ve officially launched the QUANXTA Zero series — a robot-free embodied data production platform. This is not just a hardware setup. It’s a full-stack system that connects data collection, high-fidelity synchronization, automatic cleaning, intelligent annotation, embodied model training, robot inference, and evaluation loops — all in one pipeline. The goal is very clear: close the “last mile” gap between data and models, and build scalable infrastructure for embodied AI. QUANXTA Zero comes in three setups for different capture scenarios: → G1 (UMI-VIO): dual grippers with a head-mounted rig, balancing quality, usability, and endurance → G0 (UMI-VR): full-body mobile data capture → E0 (Ego): lightweight head-mounted setup for first-person data collection In short, it’s building the data foundation for embodied intelligence. Data is what drives intelligence emergence in large models — both scale and quality matter. If the data flywheel is already spinning, real home robots might be closer than we think. @TheHumanoidHub @dolylupec