Humanoid Robots As intelligent automation systems, they are embedded within existing upstream and downstream processes, complementing and enhancing the capabilities of full-chain automation, such as production lines and warehouses. As auxiliary assistants, collaborating with humans and participating in human-led work and life, humanoid robots are learning and evolving from their masters, evolving from passive to active. As remote agents: Humans control robots through AR/VR and brain-computer interfaces, while humanoid robots act on their behalf, handling extreme tasks and meeting the human's needs. As complete agents: Humans empower the robots(AI), understanding their intentions, patterns, and preferences, autonomously planning and executing tasks, and establishing a symbiotic relationship with them. As human-machine symbionts: Humanoid robots' hands, legs, and other parts of the human body enhance the human being. As a replica of the human, embedding human memories, knowledge, experience, ego, and superego,a digital copy,within the human body,a kind of superhuman and real-world immortality.

1. Fremont shut down Model S and Model X and moved the line and the people onto Optimus. Weekly output went from a few dozen in Q2 to several hundred in August, about 10x. The year-end target is a continuous automated line above 1,000 a week, with a long-term figure around 20,000. Volume is up. The line is still unstable. 2. Most units coming off the line stay inside Tesla for testing, training, and data. Factory jobs run in supervised zones on specific tasks. What they are building now is V3. The commercial version still has a higher durability bar to clear. 3. Hands remain the hard hardware problem. Hand plus forearm has more than 100 screws and small parts, still assembled by people. The pieces are smaller and tighter than car parts. Fixtures miss alignment. Robots leave the line and go straight into rework. 4. Touch sensors on the hand have reliability issues. Tesla’s fix is a replaceable sensing glove next year, so a failed sensor does not mean swapping the whole hand. Current hands are not ready for long, continuous work. 5. Joint motors and precision gears still come from outside suppliers, many of them in China. Prototype quality is fine. Volume quality drops. Tesla is qualifying new vendors and pushing some capacity out of China. The humanoid supply base is still immature. That is an industry problem, not a Tesla-only one. 6. The real world AI has the same gap. People close to the system say the AI still fails on a wide task range and drifts in scenes it has not seen. Tesla is building a library of basic motions. Even a basic task can take days to learn. This is still a specialist, not a general worker. 7. Training data is past 500,000 hours, with a plan to double by year-end. A large part of the FSD labeling team moved to Optimus. Tesla hired collectors in camera helmets and motion-capture suits, and set up hubs in Colorado, Arizona, and Florida. Hardware is coming out. The data pipe is being built in parallel. 8. The commercial plan is lease first, sell later. Early customers will look like Tesla’s own factories and warehouses, which makes transfer easier, then fleet data feeds the next model. Same playbook as FSD: ship metal, collect miles, catch the software up later. 9. The public timeline has slipped before. In early 2025 Musk talked about 10,000 units that year and several thousand doing useful work by year-end. He later said zero Optimus units were doing useful work inside Tesla. The V3 reveal was supposed to land by mid-2026. Late September, still no demo. Factory pace and the public calendar are two different clocks. 10. Read this as scale-up friction, not collapse. A car line is now building humanoid robots. Tighter tolerances, denser assembly, a thin supplier base, and weak generalization are hitting at the same time. The hand is the physical bottleneck. The AI still needs days to learn one assigned job. Leasing instead of selling is Tesla admitting the next step is data, not a finished worker.

Two Distinct Modular Approaches > Modularization based on business scenarios: Blue Worm modularizes the humanoid robot into various robot types, ranging from humanoids to AMRs. > Modularization for ease of maintenance: O-ID modularizes components to construct the humanoid robot, allowing maintenance to be performed on-site without returning the unit to the factory. They all share a common characteristic--low-cost deployment and application. nitter.net/cyberrobooo/status/209…

Humanoid robots are going modular. Japan-based O-ID has introduced Modular, a practical humanoid robot with a wheeled base, designed to fit into existing industrial environments（manufacturing,logistics,inspection） Interestingly, the modular architecture may matter more than the specs. The robot is divided into nine replaceable modules, so a faulty arm, hand or other component can be swapped on-site instead of sending the entire robot back for repair. It stands 172 cm tall, weighs 70 kg, has 30 DoF and a 16 kg payload. The omnidirectional base is designed for factories, warehouses and inspection work. It runs Linux/ROS, while customers can choose their own AI models rather than being locked into a single system. According to Nikkei, 85% of the robot’s components by weight are made in Japan.O-ID has also signed a letter of intent with Sumitomo Electric Industries to explore the potential use of its technologies in wire harnesses for humanoid robots. O-ID plans to start selling Modular in Japan in 2027, at around $70,000.

AGIBOT just rolled off its 20,000th humanoid robot, the A3 Ultra. It will be delivered to Chimelong Group for a new embodied robotics theme park jointly developed by the two companies. In the first phase, 300+ humanoid robots will perform, interact with visitors, provide services, and support educational experiences. Thousands more are planned later. This reminds me of Disney. Disney didn’t just build theme parks. It turned characters into childhood memories, and childhood memories into decades-long consumer relationships. AGIBOT could be doing something similar with humanoid robots. Let kids grow up interacting with humanoids, making them part of their childhood memories. When this “AI-robot generation” grows up, they could become the next generation of humanoid robot consumers. So this is more than a large-scale deployment. AGIBOT may be expanding the market for humanoids while building the consumer culture around them.

IRON walks like a human. XPENG has just held a supplier conference, bringing together major suppliers across actuators, dexterous hands, sensors, AI chips and other key components. It has completed supplier audits and finalized key component suppliers. Around 85% of its suppliers overlap with XPENG’s automotive supply chain, and production scheduling is now underway. IRON’s production line started operating on September 8,Guangzhou, with more than 80% of core processes automated.

A new humanoid robot from Europe--Vesoma Its concept renders look somewhat similar to Tesla and XPENG. Vesoma says it is building a self-evolving physical agent that learns through interaction with the real world. It explores, fails, and gradually figures out how to move in ways that best suit its own body. Over time, its performance is expected to improve through continued operation. This video shows locomotion testing of Vesoma’s recently unveiled Alpha platform.

it is hard to distinguish between the humanoid robots and the human dancers. 19 Unitree H2 robots dance gracefully alongside human performers--a feat Unitree states is driven autonomously by AI. As the waltz music shifts, it truly feels like stepping into a sci-fi movie…

China Mobile just open-sourced a bridge between Physical AI models and humanoid robots. China Mobile has released Open-RAIL, an open-source real-time execution layer for VLA and World Action Models . It tackles a practical problem in humanoid robots: The Physical AI model thinks in action chunks, while the robot body needs fast, continuous motion. Open-RAIL sits in between: VLA / WAM<->Open-RAIL <-> Humanoid Robot It handles asynchronous inference, temporal alignment, and trajectory smoothing, turning low-frequency model outputs into smoother, high-frequency motion. The project says it supports 20+ VLA/WAM models and 4 types of humanoid robots（Unitree G1/AgiBot G1/China Mobile Lingxi/NAVIAI-WA2）, with tools for real-robot data collection and human intervention. In real-robot testing, Open-RAIL achieved up to 2.09× faster task execution while reducing motion jitter. Think of it as a real-time execution layer connecting robot AI models to humanoid robot bodies, making it easier to move different models across different humanoid robots.

AlphaGo defeated humans. When will humanoid robots do the same? Skild AI is cultivating the "Lionel Messi" of the humanoid robot world. Interestingly, they had the Unitree G1 engage in self-play for the equivalent of 140 years within a virtual World Cup setting. Then, in the real world, it autonomously plays soccer against humans--intercepting the ball, defending, and dribbling on the attack. Moving from the virtual to the real world eliminates the need for the linear, decade-long training required for a human soccer player; instead, the robot can gain over a century's worth of experience in the simulation in a very short time(perhaps just a few weeks) It’s soccer for now, but imagine the future possibilities if this extends to construction sites, factories, supermarkets, and households… That is what true scalability looks like.

Has Optimus v3 leaked early? These are renderings reportedly found within the APK file of the Tesla Android app. v3 has undergone a near-total redesign; it looks very human-like --from its overall form and body contours to its arms, hands, and legs. However, this does not necessarily mean the physical Optimus v3 unit will look exactly like this. We will have to wait and see if Tesla unveils it in the Q4. Given the recent flurry of reports regarding Tesla teams conducting on-site visits to supply chain partners in regions like the Yangtze River Delta, it is possible that the launch is just around the corner.

HuggingFace now hosts 1,000+ robotics “food” (datasets).🤗 Humanoid robots have a huge appetite, and this is still far from enough... we’ve only just started extracting intelligence from the physical world.

Tesla is already training Optimus. --- The factory management confirms upon inquiry: "Starting in mid-August, selected teams from Gigafactory Berlin-Brandenburg will be part of the global training for Optimus. […] This means that selected employees will wear special backpacks and helmets with cameras during their work to record their movements as a basis for Optimus." -- It appears this is something that has been underway in secret for quite some time. As of now, global shipments of humanoid robots for the first half of the year have exceeded 22,000 units. Compared to the million-unit mark, that is merely a drop in the bucket. Dear friend,everything is just getting started...

The Unitree H2 is like a roly-poly toy. He really gave it his all.