Source: Unsplash
BMW just ran a ten-month pilot with a humanoid robot working ten-hour shifts on an active assembly line. It moved over 90,000 components. It clocked 1,250 operating hours. And somewhere along the way, factory workers went from curious to comfortable. Nobody called that a UX problem. But it was.
Humanoid robots are officially on the factory floor. BMW's Figure 02 robot completed a ten-month pilot at Plant Spartanburg in South Carolina, supporting the production of more than 30,000 BMW X3 vehicles. Now BMW is rolling out AEON, a new humanoid developed by Hexagon Robotics, at its Leipzig plant in Germany, marking the first humanoid robot deployment in European automotive production. And this week, at CVPR 2026 (running June 3-7 in Denver), embodied AI and robotics are the biggest story, with robotics papers jumping from 2.9% to 6.2% of all accepted submissions. The technology is real. The deployments are real. What's not real yet is a coherent approach to designing the interaction between humans and these machines.
I've spent 8+ years designing digital products. I have built interfaces across 42 products, many of them in complex enterprise environments where humans and automated systems work side by side. And what I see playing out in humanoid robotics right now is the same mistake I've watched the enterprise software industry make for decades: deploy the capability first, design the experience somewhere down the roadmap.
"Our aim is to be a technology leader and integrate new technologies into production early. Pilot projects help us test and evolve the use of adaptive AI-enabled robots in a real-world industrial setting."
— Michael Nikolaides, Senior Vice President Production Network, BMW Group (March 2026)
What Actually Happened in Spartanburg
The BMW Spartanburg pilot is fascinating, not just for what the robot did, but for what the humans had to adapt to. Figure 02 was deployed in the body shop, a high-automation environment with experienced workers already comfortable with new technology. BMW deliberately chose this location because the workforce had "plenty of experience in using new technologies." That's not a technical decision. That's a user research decision, even if BMW didn't frame it that way.
The result? "Initial curiosity soon turned into acceptance." That sentence is a user onboarding story. Somebody had to manage that transition from novelty to normalcy, and how it was handled determined whether the pilot succeeded or fell apart. BMW noted explicitly that "the project's goals and workflows were communicated early" as a prerequisite to that acceptance. That is communication design. That is expectation management. That is UX.
The Spartanburg pilot also generated direct lessons that required revised safety concepts, additional barriers and partitions, and improved 5G coverage in the hall. These are not just technical fixes. They are interaction design decisions. Where does a human draw back when the robot approaches? How does the robot signal its next move? What physical environment is needed for both agents to share space without stress or danger? These are questions that industrial designers, interaction designers, and human factors engineers should be answering, not just mechanical engineers and software teams.
The Design Gaps Nobody Is Talking About
Traditional industrial robots have one primary UX design principle: keep humans away. Safety cages, exclusion zones, warning sounds, emergency stops. The entire human-robot interaction model is built around separation. That is relatively straightforward to design for.
Humanoid robots blow that model up completely. AEON at Leipzig rolls through the same workspace as human workers, navigates around obstacles autonomously, and works alongside people at the same stations. The interaction model is now about coexistence, not separation. And we have almost no design vocabulary for that yet.
The problems are not hypothetical. Here is what's actually going wrong in current deployments:
- Battery life forces mid-shift interruptions: Current humanoid robots run on 2-4 hours of battery. That means the robot has to leave its workstation during a shift for recharging. How does it communicate this to human coworkers? What does the handoff look like? Who takes over the task? Nobody has designed this transition yet.
- No ISO safety standard exists: Traditional industrial robots have decades of ISO standards governing their use. For dynamically balancing legged robots operating in shared human workspaces, no such standard exists. BMW is essentially designing safety protocols from scratch on live production lines.
- Connectivity drops create unpredictable behavior: Metal-dense factory environments create 180ms+ WiFi latency. SLAM navigation systems drift 20cm per 100 meters in dusty conditions. When a robot behaves unexpectedly because of a connectivity issue, what does a human worker near it do? The answer to that question is a design problem, not an engineering problem.
- Intent communication is completely unsolved: When I'm walking toward a colleague in a hallway, we do a micro-negotiation with eye contact and body language to decide who steps aside. Robots don't do that. AEON has 22 sensors but no established protocol for communicating movement intent to nearby humans in a way they can intuitively understand and respond to.
- Worker onboarding is treated as an afterthought: BMW correctly involved workers early and let them shape how the robot would be used. But this participatory design approach is the exception, not the rule. Most deployments announce the robot, run a demo, and then wonder why workers are tense or uncooperative around it.
The companies that will successfully deploy humanoid robots at scale are not the ones with the best hardware specs. They are the ones who figure out the human side of the interaction model first.
CVPR 2026 Shows Where This Is Headed
This week at CVPR 2026, the dominant theme is what researchers are calling embodied AI: systems that don't just process data but act in the physical world. The jump from 2.9% to 6.2% of accepted papers being about robotics is significant. That's not a trend. That's a field in acceleration. And the ManipArena Competition at the Embodied AI Workshop is testing robots on 20 real-world physical reasoning tasks that involve generalization, not just repeating a trained motion.
Tesla's Optimus, Figure AI's robots, Agility Robotics' Digit, and BMW's AEON all represent a wave of physical AI systems that will be working alongside humans in the next 24 to 36 months at meaningful scale. The hardware is solving the motion problem. The software is solving the perception and planning problem. But the interaction layer is still wide open.
I wrote about this exact pattern with AI-native products on Medium: when a powerful new capability arrives, the companies that win are the ones who nail the experience layer, not just the technical layer. My piece on AI-native experience design argued that the LLM is the engine, but the interface is what determines adoption. Humanoid robots are the same: the actuators and models are the engine. The interaction design is what determines whether factory workers trust, accept, and effectively collaborate with these systems.
What the Human-Robot Interaction Field Needs From Designers
Human-robot interaction (HRI) is a legitimate academic and industrial discipline. The HRI 2026 conference happened earlier this year. IEEE has working groups on this. But the practitioner design community, the people who actually ship products, is almost completely absent from this conversation.
Here is what I think designers and product teams working near this space should be thinking about right now:
Motion legibility is a design problem. A robot's movements need to communicate intent before the action completes. When a robot arm reaches for something, it should move in a way that signals "I am reaching for this specific object" rather than "I am moving in some direction." Humans do this intuitively. Robots don't, and designing legible motion paths is an interaction design challenge, not just a kinematics one.
Trust calibration needs designed checkpoints. BMW's approach of early communication and involvement was the right instinct. But for scaling beyond one factory to hundreds, you need a systematic onboarding design: what do workers learn on day one with a robot coworker? What experiences build trust over the first week? What feedback mechanisms exist for workers to report discomfort or confusion? These are product design questions.
New job roles are already emerging. "Robot fleet coordinator," "human-robot workflow designer," and "AI behavior specialist" are roles that are starting to appear. These are essentially UX roles for physical AI systems. If you are a product designer or design leader, this is worth paying attention to. The skills translate directly. Workflow design, trust design, error state design, onboarding design: all of it applies.
And at the reloadux blog, we have been writing about how AI readiness requires thinking about the interaction layer from the start, not as a retrofit. That framing applies equally to physical AI. The question is not whether your factory is ready for a robot. The question is whether your human workflows, communication patterns, and space design are ready for a coworker who does not speak, cannot read a room, and needs 5G to function properly.
The Fist Bump Problem
Boston Dynamics made a famous video of their Atlas robot giving a human a fist bump. It was delightful. It was also a carefully staged, controlled interaction designed to generate a specific emotional response: the sense that the robot is friendly, relatable, safe.
That instinct, to make robots emotionally legible to humans, is the right one. But a staged PR moment is not the same as a designed system for 200 daily human-robot interactions on a factory floor. The fist bump is easy. Designing the thousand small moments of shared workspace navigation, task handoff, error recovery, and spatial negotiation that happen over a ten-hour shift: that is the actual product design challenge.
BMW is doing real work here. Their pilot data is valuable. Their participatory design approach with workers is smart. But the industry overall is still treating interaction design as something that gets figured out after deployment, not something you design into the system from the start. That needs to change before humanoid robots move from pilots to production at scale.
The robots are ready to go to work. The question is whether we are ready to design the relationship.
Are you working on products where humans and automated physical systems share space? Or are you thinking about how interaction design applies to robotics? I'd love to hear your perspective. Drop a comment below.
Sources:
1. BMW Group: First humanoid robot introduced in Plant Leipzig — bmwgroup.com
2. CVPR 2026 Showcases How AI Is Powering the Next Era of Robotics Innovation — cvpr.thecvf.com
3. Physical AI Is Sending Humanoid Robots to Real Factory Floors in 2026 — memeburn.com
4. 5 Humanoid Robot Challenges That Block Factory Deployment in 2026 — theresarobotforthat.com
5. Humanoid Robots at Work 2026 Guide — blog.robozaps.com
6. CVPR 2026 Accepted Papers Trends — bohrium.com
7. Human-Robot Interaction in Industrial Settings — ncbi.nlm.nih.gov/pmc/articles/PMC12360944