> ## Content Index
> Fetch the complete content index at: https://news.theresarobotforthat.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Tesla Begins Building Optimus as the Humanoid Race Industrializes
- URL: https://news.theresarobotforthat.com/tesla-begins-building-optimus-as-the-humanoid-race-industrializes/
- Published: 2026-10-23T10:30:00.000Z
- Updated: 2026-10-23T10:30:00.000Z
- Description: A Japanese factory-floor first, a robot that learned a skill from one video, Unitree's best-seller claim, and a $1.4B asset-health unicorn.
- Author: Ulrich Baldauf
- Tags: Newsletter

### The Big Number

Texas Instruments now puts the value of its own chips inside a single humanoid robot at as much as $1,000, turning the humanoid build-out into a measurable semiconductor market.

## Welcome back to your Robot Briefing

This week the humanoid story stopped being about demos and started being about factories. Tesla put Optimus on a real production line, Mitsubishi Electric wired a humanoid directly into a CNC machine, and a Google model taught a robot a new skill from a single video.

The through-line is industrialization: supply chains, safety files, and unit counts are now the conversation, and the companies that can show real deployment are pulling away from the ones still shipping clips.

The 60-Second Read

Tesla started early production of Optimus on a high-volume Fremont line, with 20,000 a week still about a year out.

Mitsubishi Electric built the first machining system where a humanoid and a CNC machine coordinate through AI, no custom integration per robot.

A Google DeepMind model watched one video of a person and taught Apptronik's Apollo to sink seven cornhole shots in a row.

Unitree says its G1 is the best-selling humanoid in the world, between 10,000 and 20,000 built since 2024.

News

### Tesla Puts Optimus on a Real Production Line

**Snapshot:** Tesla has [started building](https://finance.yahoo.com/technology/ai/articles/tesla-stars-optimus-robot-production-203309413.html?ref=news.theresarobotforthat.com) its Optimus humanoid on an early high-volume line at its Fremont factory, the clearest signal yet that the robot is moving from prototype to product.

Breakdown:

Elon Musk told the third-quarter earnings call the company has begun "early production on a high-volume line in Fremont," on a line retooled from earlier vehicle work.

Reaching 20,000 units a week will take "at least a year," Musk said, with that rate targeted around the end of 2027.

Customer deliveries are also aimed at late 2027, and Musk said the first buyers are likely industrial companies with dangerous or repetitive work.

The hands remain the hard part: Tesla says parts of the [Optimus](https://theresarobotforthat.com/blog/tesla-optimus-vs-figure-03/?utm%5Fsource=newsletter&utm%5Fmedium=email&utm%5Fcampaign=internal-link) design are still changing, and quarterly profit fell 43% as the company poured money into AI and robotics.

**Takeaway:** A production line is a commitment that a demo is not, because it forces decisions on cost, yield, and serviceability. For an operations leader, the useful signal is not the 2027 date but the admission that the hands and the ramp are still unsolved, which means the first real buyers will be co-developing the robot rather than simply buying it.

News

### A Humanoid and a CNC Machine Now Talk Directly

![A Humanoid and a CNC Machine Now Talk Directly](https://storage.ghost.io/c/9c/a6/9ca6316d-1f3a-4d33-b3e6-87f231933bbb/content/images/2026/10/story-2026-10-23-1.jpg)

Image source: Mitsubishi Electric

**Snapshot:** Mitsubishi Electric [developed](https://www.mitsubishielectric.com/en/pr/2026/1021%5Frd/?ref=news.theresarobotforthat.com) what it calls the industry's first machining system in which a humanoid robot and a CNC machine tool coordinate through physical AI, letting the robot take over operator tasks around the machine.

Breakdown:

An AI machining task planner takes an operator's instructions in plain language, generates the work, and splits it between the CNC machine and the humanoid.

The humanoid handles setup, cleaning, inspection, and removing and transporting finished parts, the manual work that normally keeps a person at the machine.

The CNC machine itself runs physical AI and exchanges status with the robot in natural language, so there is no dedicated integration to build for each robot model.

Mitsubishi Electric will demonstrate the system at JIMTOF 2026 in Tokyo from October 26, and frames it as a step toward autonomous, unmanned machining.

**Takeaway:** The quiet breakthrough here is the interface, not the robot. Making the machine and the humanoid speak the same language removes the per-robot integration cost that has made humanoids expensive to slot into existing lines, which is exactly the barrier that keeps most factory pilots from scaling.

⏱️ The Compliance Clock

With Tesla building Optimus on a Fremont line and more US plants trialing humanoids beside staff, the governing references for a US site are ANSI/RIA R15.06, the American safety standard for industrial robots, and OSHA, which enforces worker safety through its General Duty Clause. R15.06 raises what an operator has to document before a humanoid shares floor space with people: a task-based risk assessment and validated safety functions, not a demo reel. For any US pilot, that file is what an inspector and an insurer will ask to see first.

News

### A Google Model Taught a Robot a Skill From One Video

**Snapshot:** Google DeepMind [showed](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/?ref=news.theresarobotforthat.com) that its new Gemini 4 Argon model can watch a single video of a person, work out the motion, and transfer the skill to Apptronik's Apollo humanoid.

Breakdown:

In an internal test, Argon watched raw video of a human playing cornhole, mapped the space, and calculated the whole-body joint movements needed.

The model then wrote its own simulation code, trained the throw, and handed it to Apollo, which sank seven consecutive shots in a real trial.

DeepMind calls the method "Video-to-Sim-to-Real" and says it compresses weeks of manual simulation modeling into a few hours.

Argon also posts an 87% score on the Embodied Reasoning benchmark, which tests a model's grasp of physical and spatial problems.

**Takeaway:** The cost that has slowed humanoid deployment is not hardware, it is teaching each robot each new task. A pipeline that turns ordinary video into a trained skill attacks that cost directly, and it points toward a future where a new line task is taught by filming a worker rather than by hand-coding a program.

News

### Unitree Claims the Best-Selling Humanoid as Its Founder Calls a Turning Point

![Bar chart comparing the 8-hour runtime of Unitree's H2-X humanoid against the 3-hour runtime of the standard H2, alongside a bold callout highlighting up to 20,000 G1 units built since 2024.](https://res.cloudinary.com/duuood1js/image/upload/f_webp,q_auto/v1792722420/theresarobotforthat/C267-infographic-2026-10-23.png)

Image Source: There's A Robot For That

**Snapshot:** Unitree founder Wang Xingxing [told](https://www.globaltimes.cn/page/202610/1372791.shtml?ref=news.theresarobotforthat.com) the World Physical AI Conference in Hangzhou that the company's G1 is the world's best-selling humanoid, with roughly 10,000 to 20,000 units built since its 2024 launch.

Breakdown:

Wang put cumulative G1 production at between 10,000 and 20,000 units since May 2024, a self-reported figure that still outpaces most Western rivals' deployment counts.

He argued physical AI is showing its first real signs of a breakthrough, which he expects to land within one to three years.

A day earlier Unitree unveiled the H2-X "Kung Fu Legend Guardian," a security and rescue humanoid with a claimed eight-hour runtime, up from three hours on the standard H2.

In Unitree's demo video the H2-X shielded a person from an attack and lifted the front of a car, the security and rescue scenarios the company is pitching it for.

**Takeaway:** Unit counts are becoming the real scoreboard, and Unitree is winning it on price and volume while US makers debate specs. For a buyer, the caution is that "best-selling" counts research labs, universities, and demo units, so a shipped G1 is not the same as a humanoid doing paid work on a line.

▶ Operator's Playbook

Before you let any humanoid near staff this quarter, ask the vendor for its task-based risk assessment and the validated safety functions under ISO 10218 or ANSI/RIA R15.06, in writing. If the answer is a demo video or a promise to "handle compliance later," treat the pilot as a research project, not a deployment, and price it that way.

[Get the Toolkit →](https://theresarobotforthat.com/safety-compliance-kit/?utm%5Fsource=newsletter&utm%5Fmedium=email&utm%5Fcampaign=playbook)

### Other Top Robot Stories

**Tractian** [raised $130 million](https://www.prnewswire.com/news-releases/tractians-130m-series-d-values-the-company-at-1-4b-to-scale-physical-ai-for-asset-heavy-industries-302917313.html?ref=news.theresarobotforthat.com) in a Series D led by Sapphire Ventures that values the asset-health company at $1.4 billion, with more than 2,000 facilities now running its physical-AI monitoring across pharma, food, energy, and data centers.

**NVIDIA** [expanded its Isaac platform](https://nvidianews.nvidia.com/news/nvidia-advances-isaac-platform-for-general-purpose-robotics-with-new-reference-humanoid-frameworks-simulation-pipelines-and-data?ref=news.theresarobotforthat.com) at GTC Berlin with a new Isaac OS runtime and general availability of Isaac Lab 3.0, alongside Agile Robots' Agile ONE SN, a 1.40-meter reference humanoid for researchers due in early 2027.

**ESAB** [launched ROBBI AI](https://www.businesswire.com/news/home/20261022098475/en/ESAB-ROBBI-AI-Cobot-for-MIG-and-TIG-Welding-Automation-Brings-Physical-AI-to-the-Shop-Floor?ref=news.theresarobotforthat.com) , a welding cobot that uses camera and laser sensing to generate its own weld path and adjust in real time as a joint gap opens or closes, aimed at the high-mix, low-volume jobs that are hardest to staff.

**Teradyne Robotics** [partnered with Sereact](https://www.universal-robots.com/news-and-media/news-center/teradyne-robotics-and-sereact-partner-to-scale-physical-ai-solutions-for-retail-and-logistics?ref=news.theresarobotforthat.com) to run the startup's Cortex vision-language-action model on Universal Robots arms, with hundreds of UR robots in dual-arm systems already handling returns and piece-picking at live European logistics sites.

**FORT Robotics** [secured about $20 million](https://www.fortrobotics.com/news/fort-robotics-secures-approximately-20-million-in-additional-funding-led-by-tiger-global-and-mark-cuban-companies?ref=news.theresarobotforthat.com) from Tiger Global and Mark Cuban Companies as it heads for a Nasdaq listing through a SPAC merger, pitching its "trust layer" safety software, now running on more than 19,500 machines for over 600 customers, as the independent safety layer physical AI will need.

Compliance toolkit

The EU Machinery Regulation applies from January 20, 2027.

The Robot Safety Documentation Toolkit gives your team all seven compliance documents, ready to fill in.

[Get the toolkit →](https://theresarobotforthat.com/safety-compliance-kit/?utm%5Fsource=newsletter&utm%5Fmedium=email&utm%5Fcampaign=promo-strip)

**🤖 Your robotics thought for today:**

The best robotics news this week was not a backflip or a faster chip. It was a machine and a robot agreeing on how to talk to each other. Progress usually looks less like a breakthrough and more like two systems that finally stopped needing a translator.

Enjoy your weekend,  
Uli