Physical AI
Siemens Puts Humanoid's HMND 01 on the Factory Floor, Hits 60 Tote Moves an Hour
Siemens said Humanoid's HMND 01 Alpha wheeled robot completed autonomous logistics work inside its Erlangen electronics factory at 60 tote moves per hour, with uptime above eight hours and pick-and-place success above 90 percent. The trial matters because it ties a young humanoid platform to Siemens' industrial software stack and NVIDIA's physical AI tooling, turning a flashy robot demo into an operations story.
Siemens says Humanoid's HMND 01 Alpha wheeled robot has completed autonomous logistics work inside its electronics factory in Erlangen at 60 tote moves per hour, with more than eight hours of uptime and over 90 percent autonomous pick-and-place success. Those are the kind of operating numbers the humanoid sector has been short on, because most announcements still revolve around flashy videos, lab trials, or future production promises.
The April 16 announcement matters because it combines three pieces that rarely show up together in one deployment story: a live factory environment, an industrial software backbone through Siemens Xcelerator, and a modern robot AI stack built on NVIDIA Jetson Thor, Isaac Sim, and Isaac Lab. For biped and humanoid robotics, that is a stronger signal than another funding round. It suggests the market is starting to care less about whether a robot can move like a person and more about whether it can clear a shift, hit throughput, and fit into plant operations without becoming its own integration project.
AI-generated image
Editorial illustration of factory intralogistics and automation infrastructure. Source: OpenAI image generation for biped.news.
Key Stats
60/hr
Tote Moves
8+ hrs
Reported Uptime
90%+
Autonomous Success
7 mos
Prototype Cycle
How the Erlangen Trial Worked
The use case was not glamorous, which is exactly why it matters. Siemens and Humanoid focused on tote-to-conveyor destacking inside the Siemens Electronics Factory in Erlangen, Germany. In practical terms, the robot picked totes from storage stacks, transported them through the logistics flow, and placed them at a pickup point for human operators. That is the kind of repetitive intralogistics job factories want automated first because it burns labor hours, creates bottlenecks, and does not require a robot to solve every manipulation problem at once.
According to Siemens and coverage of the earlier proof of concept, the project ran in two phases. First, Humanoid built and tuned the system off-site with a physical twin and simulation workflows. Then the companies moved into a two-week live deployment in Erlangen to measure real operating performance. The reported targets were unusually concrete for this sector: handling two tote sizes, sustaining continuous autonomous task execution, delivering more than eight hours of uptime, and keeping both overall and autonomous pick-and-place success above 90 percent.
What Siemens announced on April 16 is the step beyond a one-off robotics pilot. The company framed the result as evidence that humanoid systems can become collaborative assets on the shop floor, exchanging data with production software, working alongside autonomous guided vehicles, and adapting to changing conditions rather than operating as a disconnected demo. That distinction is the whole story. A robot that can move a tote in isolation is interesting. A robot that can plug into a real factory's control, communications, and workflow stack starts to look commercial.
What the robot actually did
Autonomous tote picking, transport, and placement in an intralogistics workflow, with humans still receiving material at the line-side handoff point.
Why factories care
The task is repetitive, easy to measure, and expensive to staff over multiple shifts, which makes it one of the first sensible entry points for humanoid-class robots.
Key Insight
The headline number is not that HMND 01 looks human. It is that Siemens published throughput, uptime, and success-rate metrics from a live plant. That moves the conversation from robotics theater to factory economics.
Under the Hood: Why HMND 01 Is Built More Like Factory Equipment Than a Robot Demo
Humanoid, founded in 2024 by Artem Sokolov, has moved unusually fast. Siemens says the company used simulation-first design and NVIDIA's physical AI toolchain to cut prototype development from a typical 18 to 24 months down to seven months. That is important because most humanoid companies are still spending heavily to iterate hardware that may not survive real industrial duty cycles.
The deployed system in Erlangen was the HMND 01 Alpha wheeled variant, not a pure biped. That choice is pragmatic. On Humanoid's own published specs, the platform stands 220 centimeters tall, weighs 300 kilograms, carries a 15 kilogram payload, moves at up to 2 meters per second, and runs for an average of four hours on its internal power system. The platform uses 29 degrees of freedom excluding end effectors, plus RGB cameras, depth sensors, force-torque sensing, and haptic feedback. Humanoid also offers interchangeable end effectors, including a five-finger hand and a simpler parallel gripper.
That hardware choice says a lot about where industrial physical AI is heading. For logistics, a wheeled humanoid often makes more sense than a full biped because it trades some stair-climbing versatility for better stability, lower control complexity, and more predictable energy use. Factories are mostly flat, structured spaces. The market reward there goes to the robot that can complete shifts, avoid safety incidents, and clear work cells, not the robot that looks most anthropomorphic in a teaser reel.
AI-generated image
Editorial illustration of the digital twin and edge-compute stack behind industrial physical AI. Source: OpenAI image generation for biped.news.
| Spec | HMND 01 Alpha Wheeled | Agility Digit | Figure 03 |
|---|---|---|---|
| Mobility | Omnidirectional wheeled base | Bipedal walking | Bipedal walking |
| Height | 2.20 m | 1.75 m | Not publicly fixed |
| Weight | 300 kg | ~65 kg | Not publicly fixed |
| Payload | 15 kg | ~16 kg | Not publicly disclosed |
| Runtime | 4 hrs average, 8+ hrs reported site uptime | Shift-based RaaS deployments | Multi-hour factory and home demos |
| Commercial status | Pilot / early deployment | Commercial deployment | Pilot deployment |
Technical Context
• Edge compute: Siemens said HMND 01 uses NVIDIA Jetson Thor for onboard AI inference.
• Training stack: NVIDIA Isaac Sim and Isaac Lab were used for simulation and reinforcement-learning workflows.
• Factory integration: Siemens Xcelerator provides digital twin tooling, PLC-robot interfaces, fleet management, and industrial networking.
• Robot strategy: The wheeled variant lowers mobility risk in flat industrial environments while keeping human-scale reach and manipulation.
Who's Building This, and Why the Siemens Plus NVIDIA Stack Changes the Story
Humanoid is still young enough that many manufacturing executives will not know the name yet. The company launched publicly in 2025, says it has more than 200 engineers and researchers, and has offices in London, Cambridge, Massachusetts, and Burnaby, British Columbia. On its own, that profile would make it one of many promising robotics startups trying to turn prototypes into contracts.
The difference here is the partner set. Siemens is not just a logo on a slide. It is effectively acting as industrial validator, providing a real factory environment and the software layers required to connect a robot to production infrastructure. NVIDIA contributes the physical AI stack that much of the sector now sees as standard for simulation, policy training, and edge inference. Together, the arrangement gives Humanoid something more valuable than buzz, a path into the procurement language of manufacturers.
• Siemens: Provides the factory environment, Xcelerator software stack, controls interfaces, digital twin tools, and the credibility of a global industrial automation supplier.
• NVIDIA: Supplies the simulation and compute backbone that lets Humanoid compress iteration cycles and train behavior before hardware hits the floor.
• Humanoid: Brings the robot platform and its KinetIQ orchestration framework, with both wheeled and bipedal product variants aimed at industrial use cases.
This is also a European story, which makes it strategically interesting. Much of the humanoid coverage over the past six months has been dominated by U.S. leaders like Figure and Agility or Chinese scale stories from UBTECH, Unitree, and Agibot. Humanoid gives Europe a contender that is not arguing from concept art or distant roadmaps. It is arguing from a specific workflow, a specific factory, and a specific performance envelope.
200+
Humanoid team members
2
Published HMND 01 form factors
Apr 20
Hannover Messe opens
What This Means for Physical AI, Factory ROI, and the Shape of the Humanoid Market
The humanoid market has been split between two narratives. One says the winning robots will look and move as much like humans as possible. The other says the winners will be whatever machine shape fits the economics of the task. HMND 01 leans hard toward the second view. It keeps the upper-body advantages of a humanoid platform while using wheels where factories are already optimized for flat, repeatable movement. That may not be as cinematic as a fully biped robot crossing rough terrain, but it is a better fit for intralogistics.
There is also a business lesson in the way Siemens framed the project. The company did not pitch the robot as a labor replacement fantasy. It described the value in terms of data exchange, synchronized workflows, fleet coordination, and adaptive behavior inside the wider factory system. In other words, the real product is not the robot alone. It is the robot plus integration stack. That favors players who can pair hardware with simulation, safety, controls, and deployment tooling, which is why NVIDIA and Siemens keep showing up around the same projects.
For competitors, the trial raises the bar. Companies can no longer rely on announcing that a robot walked, sorted one object, or completed a short lab benchmark. Manufacturers will want to know shift uptime, success rate, throughput, handoff behavior, and how many plant systems the robot can plug into without months of custom work. The sector is starting to borrow the metrics culture of warehouse automation and industrial software, and that is healthy.
AI-generated image
Editorial illustration of the operations data layer that determines whether a robot deployment is useful or just impressive. Source: OpenAI image generation for biped.news.
Why this story beats a funding headline
Investors can price hope. Factories do not. A deployment story with throughput and uptime numbers is more valuable than another valuation headline because it tells buyers where the robot sits on the path from pilot to line item in a plant budget.
What's Coming Next
The timing of the announcement is not accidental. Hannover Messe 2026 opens on April 20, and NVIDIA, Siemens, and other industrial AI vendors are using the event to push the idea that physical AI is moving from research to deployment. That means the next useful signals to watch are not polished keynote clips. They are customer references, follow-on sites, and whether Siemens expands the scope from one intralogistics workflow to several.
For Humanoid specifically, the next 12 months will revolve around repeatability. Can the company take the HMND 01 system from one validated use case in Erlangen to several factories, several workflows, and eventually larger fleet counts? If the answer is yes, Europe will have a serious new player in industrial humanoids. If the answer is no, this will still stand as a revealing case study in what the market now expects from robot companies claiming they are ready for production work.
Frequently Asked Questions
Is HMND 01 really a humanoid if the deployed version uses wheels?
Yes, in the broad industry sense. Humanoid is building human-scale upper-body manipulation systems with both wheeled and bipedal lower-body options. The deployed Alpha model uses wheels because flat industrial sites reward stability and efficiency more than fully human-like locomotion.
What did Siemens actually prove in this trial?
Siemens said the robot met its target metrics in a live factory workflow, including 60 tote moves per hour, more than eight hours of uptime, and autonomous pick-and-place success above 90 percent. That does not prove full-scale fleet deployment yet, but it does show the system can complete a measurable intralogistics task inside an operating plant.
Why is NVIDIA involved in so many physical AI announcements?
NVIDIA now provides a common stack for robot simulation, policy training, and edge inference. For startups, that shortens development cycles. For manufacturers, it creates a more standardized path from digital twin and simulation work to on-site deployment.
How does HMND 01 compare with Digit or Figure?
Digit and Figure are more visibly associated with bipedal factory work, while HMND 01's deployed version uses an omnidirectional wheeled base. That makes HMND 01 less general in rough terrain, but potentially better suited to flat indoor logistics where stability, uptime, and control simplicity matter more than stair-climbing.
The 12-Month Outlook
April's Siemens announcement does not mean the humanoid labor shortage story is solved. It does mean the conversation has advanced. One more company has crossed from concept positioning into measurable factory operations, and it did so with partners that buyers already trust.
The next phase for the sector will be won by teams that can repeat this kind of deployment across multiple sites and tasks while keeping integration costs under control. That is harder than building a good demo, but it is finally the problem the market seems ready to pay for.
The Bottom Line: Siemens gave Humanoid something every young robot company wants and very few have, a real factory benchmark that ties physical AI hype to throughput, uptime, and plant integration.
If more announcements this year start looking like this one, with concrete metrics and clear workflow boundaries, the humanoid market will be in much better shape than it was at the start of 2026.