For two years the story of humanoid robots in manufacturing was told mostly in demo videos. In October 2026 it started to be told in production plans, commercial agreements and regulatory decisions. Within ten days, Tesla said Optimus production had begun on its high-volume line in Fremont, AGCO signed a pilot agreement with Agility Robotics, Schaeffler opened a factory where humanoids assemble components for other humanoids, Audi started a parts-sequencing proof of concept with mobile humanoids, and South Korea cleared 20 humanoids to work in spaces shared with people. Read carefully, these announcements say less about robots replacing workers next year and more about where the real bottlenecks now sit: reliability, integration and safety certification.
What actually happened in October 2026
Tesla: early Optimus production, with a long ramp
In its third-quarter 2026 update, published on 21 October, Tesla wrote that Optimus production had begun in Fremont on its high-volume lines and that "the bots coming off this line are the first production-quality bots that we have manufactured." It added that it would keep iterating on the hardware and manufacturing process.
On the earnings call, Elon Musk called this "early production" and set expectations carefully. He described the ramp as an S-curve whose first part is shallow and "somewhat bumpy", said it would probably take "at least a year, maybe a little longer" to reach an annualized rate of 20,000 units a week, and said that parts of the design were still changing because the robot must be "safe and reliable and has long duration." He expected those issues to be solved around the end of 2027, which is when he said Tesla expects to deliver Tesla Optimus robots to companies outside Tesla. TechCrunch noted that the same quarter's net income fell 43% year over year, partly because of spending on AI and Optimus.
Boston Dynamics and Hyundai: a training ground before the line
Boston Dynamics opened its Robotics Metaplant Application Center (RMAC) at Hyundai Motor Group Metaplant America in Georgia on 21 September 2026. Atlas robots there are being trained on parts sequencing and material handling, with real-world data fed back into the AI model. According to Hyundai Motor Group, Atlas is slated to begin parts-sequencing work at the plant in 2028 and to expand into parts assembly from 2030. The Group plans 25,000 Atlas units across Hyundai and Kia plants over the next few years and a US facility able to build 30,000 robots a year; RMAC itself is due to move into a building about ten times larger in 2027.
Labour relations are part of the plan. The Seoul Economic Daily reported in April 2026 that Hyundai had told a lawmaker it would not deploy Atlas at its domestic Korean plants, after its union said no robot could enter domestic production sites without a labour-management agreement.
AGCO and Agility Robotics: a commercial pilot for 2027
On 20 October AGCO and Agility Robotics signed a commercial agreement for a US humanoid robot factory pilot, expected to begin in 2027 at AGCO's Hesston, Kansas, plant. Digit will start with parts sorting and stacking and the placement of totes for storage, and will be managed through Agility Arc, Agility's cloud platform for robot fleets. AGCO's Director of Manufacturing Technology, Thomas Enghof, framed the robots as "teammates" and stressed "strict safety governance". Agility says Digit has more than 65,000 hours of real-world operation and lists Toyota, GXO and Schaeffler among its customers.
Schaeffler: humanoids making humanoids
Also on 20 October, Schaeffler opened a roughly 7,500 square metre "lighthouse factory" in Taicang, China. At its centre is an assembly line for humanoid robot components, including linear and rotary actuators, where six humanoids work at five stations staging, sorting, joining, inspecting, labelling and packaging parts. A digital twin mirrors the line in real time, and a training and data acquisition centre teaches humanoids tasks such as visual inspection, packaging, container moving and line-side replenishment, using environments modelled on Schaeffler's approximately 100 plants. Schaeffler says its incoming orders for humanoid robotics have risen to some €400 million.
Audi: the hardest picking problem in the plant
On 21 October at NVIDIA GTC Berlin, Agile Robots, Audi, Dassault Systèmes and NVIDIA announced a proof of concept for parts sequencing at Audi's Ingolstadt plant. Audi says around one million picks a day put the right parts at the right workstation in the right order. In Audi's Production Lab, Agile Robots' mobile humanoids, dual-arm upper bodies on a mobile base, pick parts from containers and place them in sequence. Before any deployment, the partners are building a virtual twin of the logistics environment to test feasibility, timing, ergonomics and safety. Audi's board member for production and logistics, Gerd Walker, said picking within sequencing "has remained one of the toughest automation challenges in automotive logistics" for decades.
Korea: removing the fence, with conditions
On 13 October South Korea's Ministry of Trade and Industry granted a regulatory sandbox exemption that lets robot maker Robros deploy 20 bipedal humanoids at Nepes, Sungwoo Hitech, HD Hyundai Samho and Delivery Lab, in spaces shared with or next to human workers. Until now, Korean rules treated industrial robots as stationary equipment that required safety fences. The exemption is conditional on risk assessments and safety measures, and the data gathered will feed future safety standards. The same report cited BMW's deployment of Figure 02 at Spartanburg, which handled more than 90,000 parts over roughly ten months, and Shanghai's goal of 8,000 humanoids in factories by 2028.
The pattern: logistics first, assembly later
Put side by side, the first jobs are strikingly similar: sorting, stacking, totes, kitting and parts sequencing. Hyundai starts Atlas on sequencing in 2028 and moves to assembly in 2030. AGCO starts Digit on sorting and storage. Audi's first use case is sequencing. Even Schaeffler's humanoids are joining and inspecting components on a line built for them.
This matches what buyers say. In an August 2026 survey of a select group of senior facilities and supply chain leaders, Altman Solon found that case and item picking topped the list of tasks they would hand to a general-purpose robot in the next two years, with palletizing and internal material movement close behind. The same leaders estimated that manual work fills about 61% of productive floor hours today and expected that to fall to about 35% within five years. Altman Solon concluded that the form factor "barely matters" to these buyers: capability and payback drive the decision. Before a pilot, they want proof of uptime, safety and throughput from a comparable live operator, a clear payback, easy integration with existing layouts and pricing models in which vendors share the risk.
Timelines tell the rest of the story. Most dates attached to these announcements are 2027 (AGCO's pilot, Tesla's external deliveries, RMAC's expansion), 2028 (Atlas sequencing, Shanghai's target) or 2030 (Atlas assembly). October 2026 is the start of industrial validation, not the end of it.
Why safety certification is the real gate
The most important document of the season is not a press release. It is a committee draft. ISO/CD 25785-1 sets out safety requirements for industrial mobile robots with "actively controlled stability", robots that need active control to stay balanced and could become unstable without power, including bipedal and quadrupedal machines. Its comment period closed on 8 July 2026 and it is still under development; a separate Part 2 will cover the integration of applications. Until a humanoid robot safety standard is published, integrators have to argue safety case by case.
A Siemens research paper on arXiv from August 2026 explains why this is hard. Classical machine safety assumes that cutting power produces a safe stop. For a walking humanoid, removing power causes an uncontrolled fall, so the safe state is a balanced standstill that depends on the robot's own control software. The authors call this the "fail-passive gap". They built a certified external safety chain (light curtain, emergency stop, fail-safe PLC and wireless PROFIsafe) around a Unitree G1 pick-and-place cell measuring 3 by 1.5 metres. When they cut communication, the robot reached a stable balanced standstill in 0.5 to 1.3 seconds in every trial, but they are explicit that the robot-side reaction chain cannot yet be certified end to end.
Korea's sandbox, Audi's virtual twin and AGCO's "strict safety governance" are all responses to the same gap. The robots can move; the industry still lacks an agreed way to prove they will stop safely.
My view: treat a humanoid like a new production system, not a new gadget
I have spent much of my career moving critical systems from pilot to production, and I read these announcements the way I would read any platform migration: by asking what is committed, what is measured and who carries the risk. Here is the playbook I would use for a humanoid pilot.
- Start with the task, not the form factor. Choose a sequencing, kitting or tote-handling job with clear volumes and quality rules. If a fixed arm or an autonomous mobile robot can do it, buy that instead.
- Validate in a twin, then in a lab, then on the line. Audi's virtual twin, Hyundai's RMAC and Schaeffler's training centre all follow this sequence. Budget time for each stage.
- Measure before you automate. Record baseline cycle time, error rates, injuries and staffing for the target task so that any vendor claim has something to be compared against.
- Demand real-world numbers. Ask for mean time between failures, uptime per shift, interventions per hour and protective-stop rates from comparable live sites, not clips. This is the same discipline I argued for with software agents in AI agent benchmarks in 2026.
- Write the safety case early. Know which standards you are mapping to, how the robot stops when power, network or software fails, and how separation distances are calculated. Track ISO 25785-1 as it moves forward.
- Plan the fleet and data layer. Fleet software such as Agility Arc, plant systems and the data loop that retrains the robot are where integration risk lives. Decide who owns the operational data and where it is processed.
- Bring the workforce in from day one. Hyundai's decision on its Korean plants shows that labour agreements can decide where robots go. Frame the pilot around ergonomics and hard-to-staff jobs, and share results openly.
- Prefer contracts that share risk. Outcome-based or service pricing, clear exit terms and agreed performance thresholds protect you while the technology matures.
The layers underneath matter too. In Humanoid whole-body control in 2026 I covered the research that keeps these robots balanced while they carry and react, and in Physical AI in 2026 the vision-language-action models that give them their task intelligence. Factory pilots are where both layers are finally tested against a shift schedule.
Closing thought
October 2026 did not prove that humanoids are ready to run factories. It proved that serious manufacturers are now willing to test them under real production conditions, with dates, contracts and regulators attached. The winners over the next three years will not be the companies with the most impressive video. They will be the ones that can show, with numbers, that a robot works a full shift, integrates with existing lines and stops safely every time.
References
- Q3 2026 Update, Tesla, 21 October 2026
- Tesla (TSLA) Q3 2026 Earnings Call Transcript, The Motley Fool, 22 October 2026
- Tesla starts Optimus robot production in California as profit falls 43%, Kirsten Korosec and Sean O'Kane, TechCrunch, 21 October 2026
- Boston Dynamics Opens Robotics Metaplant Application Center to Train Humanoid Robots for Manufacturing Tasks, Boston Dynamics, 21 September 2026
- Inside the Robotics Race: Atlas Learns the Real World, Hyundai Motor Group
- Hyundai Motor to Break Ground on Ulsan Plant Rebuild Next Year, Rules Out Atlas Deployment, Kim Woo-bo, Seoul Economic Daily, 5 April 2026
- AGCO and Agility Robotics Advance Smart Manufacturing with Humanoid Robot Pilot in Hesston, Kansas, Facility, AGCO, 20 October 2026
- Humanoids making humanoids: Schaeffler opens lighthouse factory in Taicang, Schaeffler, 20 October 2026
- Agile Robots and Audi to transform automotive logistics with mobile humanoids, supported by Dassault Systèmes and NVIDIA, Audi, 21 October 2026
- Korea Clears Safety Rules to Put 20 Humanoid Robots on Factory Floors, Lee Jung-hoon, Seoul Economic Daily, 14 October 2026
- Humanoid Robots on the Factory Floor: Convinced, but Waiting for Proof, Altman Solon, August 2026
- ISO/CD 25785-1: Robotics — Safety requirements for dynamically stable industrial mobile robots — Part 1: Robots, ISO/TC 299
- Toward Certified Functional Safety for Industrial Humanoid Robots: The Fail-Passive Gap and a Feasibility Study, Caiwu Ding, Tao Cui, Lingyun Wang and Chengtao Wen, Siemens, arXiv, August 2026