OpenAI Split Its Robotics Team. Recruit the Half With No arXiv Trail
OpenAI moved its robotics researchers to Mark Chen's org and its hardware staff to Applied. Here is how to find the controls engineers who never publish.
If you hire robotics engineers, OpenAI's Nov 13, 2026 reorg looks like a talent release, and the obvious move is to message the researchers. That's the wrong half. The researchers moved into Mark Chen's core research org, and anyone can find them on arXiv and CoRL author lists. The applied robotics and hardware staff moved into Greg Brockman's Applied org. They, along with the lab's data collectors, publish nothing, and they're the people a robotics startup has a real chance of hiring.
What actually happened to OpenAI's robotics team?
OpenAI split its robotics group in two. Researchers now report to Chief Research Officer Mark Chen, and the applied robotics and hardware teams moved into the larger Applied organization under Greg Brockman. No layoffs were reported, and OpenAI didn't disclose headcount.
People are arguing over the "OpenAI robotics team disbanded" headline. Michelle Fradin, who works on robots at OpenAI, called it misleading and said robotics research had been merged into the main research team. She has a point. Still, the details around the reorg tell a recruiter plenty:
- CoRL booth #208 sat empty, with no OpenAI logo, while a few former team members still drew a crowd there.
- Postings stayed live. As of Nov 14, OpenAI was still hiring for on-board computing for robots, control software, and motors and actuators.
- The number of openings went up. Per Business Insider, OpenAI's robotics openings grew from 11 in May to 27, covering machine learning, robot control, electronic engineering, actuators, firmware, and data collection.
- The hardware roles are physical. The Motors & Actuators posting asks for stator process, tooling/fixture, or prototype manufacturing engineers, on site five days a week in San Francisco.
- The leadership bet is real. OpenAI hired Caitlin Kalinowski, a former Meta hardware executive, for robotics in 2024.
So this is a reorg of a team that is still hiring. It is not a shutdown. Anyone you approach still has a job and is well paid. Treat this as a poaching pitch, not a rescue.
Why shouldn't you chase the robotics researchers?
The researchers just landed in the most prestigious seat in the company, and they're the most visible people in the reorg. Their names sit on arXiv papers and CoRL author lists, so every other recruiter can find them as easily as you can.
Look at what the move means for them. Guanya Shi of CMU told Fortune that OpenAI now sees robotics as "no longer a special problem." If you're a robot learning researcher, that's good news. Your work now sits in the same org as the frontier models, which is usually the job researchers wanted in the first place. That's not the moment most people leave.
The applied side is in a different position. Brockman's Applied org is betting that general models like GPT-6 Astra can run robots directly. If the model does the robotics, the actuators, firmware, and control loops become the parts that just have to work and cost less. In a model-first company, that makes the hardware layer a cost centre, and engineers notice when their work is treated that way.
The researchers got promoted. The hardware engineers got reassigned to an org that thinks the model can do their job.
How big is the hardware talent pool compared to research?
In Refolk's index of US professional profiles, 380 people have "robotics" in their headline along with a controls, hardware, embedded, mechatronics, or actuator title. Another 247 have a research scientist or research engineer title. Hardware outnumbers research 1.54 to 1, and the two groups work at almost entirely different employers.
| Segment (US, headline keyword match) | Profiles | Top employers in sample |
|---|---|---|
| "robotics" + Research Scientist/Engineer | 247 | Meta, Google DeepMind, Apple |
| "robotics" + Controls/Hardware/Embedded/Mechatronics/Actuator | 380 | Ford, GM, Verdant, Serve Robotics |
| Hardware-to-research ratio | 1.54 : 1 | n/a |
| OpenAI open robotics roles, May to Sept 2026 | 11 to 27 | Business Insider |
The employer lists come from 25-profile samples, so read them as patterns, not rankings. Still, the split is clear. Research titles cluster at Meta, Google DeepMind, and Apple. Hardware and controls titles cluster at Ford and GM, then at robotics companies like Verdant, Serve, Bedrock, Icarus, and Plus One Robotics.
Why Ford and GM top the hardware list
Ford and GM top the list because many controls and embedded engineers describe themselves by the industry they work in, not by the word "robotics." Their headlines say automotive or EV, not robot. The skills are close to what a humanoid needs. Figure, for example, builds its motors and stators in-house, which is motor manufacturing work an automotive engineer will recognize.
That makes 380 a floor, not a ceiling. If you want to source robotics hardware engineers, the robotics keyword filter has already thrown out most of them. Search automotive, EV, and motor-manufacturing backgrounds, then check for robotics-adjacent work in their code or patents.
Where do robotics engineers who don't publish leave a trail?
They leave it in code, patents, conference floors, and employer histories instead of author lists. Each of these is a separate search, and together they replace the arXiv scrape that only finds researchers.
- Public code. Commits to robot drivers, controller code, and simulation assets are written by the engineers you want, not by paper authors.
- Patents. Actuator and stator patents name inventors who never write a conference paper.
- Conference staffing. At CoRL, former OpenAI team members drew a crowd at an empty booth. The people running demos, staffing booths, and fixing robots between sessions are usually applied engineers.
- Past employers. Figure is the obvious humanoid alumni pool. The hardware cluster from Refolk's index (Ford, GM, Verdant, Serve, Bedrock, Icarus, Plus One Robotics) is the less crowded one.
- The job posting's own vocabulary. Search the titles OpenAI is hiring for: stator process engineer, tooling/fixture engineer, prototype manufacturing engineer, firmware engineer. Anyone who holds those titles today is a candidate for the same work at your company.
Doing this by hand means jumping between GitHub, LinkedIn, patent databases, and conference agendas, then merging the results in a spreadsheet. Refolk handles that step. You describe the engineer in plain English, and it searches GitHub, LinkedIn, and the open web and returns a ranked shortlist, so a commit history and a LinkedIn headline end up on the same profile.
Why does searching "humanoid" return the wrong people?
On a profile, "humanoid" in the headline mostly means a founder. In Refolk's index the keyword matches 4,624 US profiles across all titles, but the top titles in the sample are founder, CEO, and CTO at companies like Persona AI and Figure. That number is noisy, because the keyword fuzzy-matches unrelated companies, so use it to judge direction, not size.
Executives use "humanoid" because it names the category their company sells. The engineers building the legs and hands don't, because their job is a subsystem, not a category. If you want to recruit controls engineers for humanoid robots, search for the subsystem: actuators, motor control, firmware. Then add a past employer.
That's also how to write a Refolk query. Name the subsystem and the previous employer, not the buzzword, because that matches how hardware engineers describe their own work.
Who is the most overlooked group in OpenAI's robotics org?
It's the teleoperation data collectors. OpenAI built a robotics lab with around 100 data collectors who teach a robotic arm to do household tasks. Teleoperation means a human remotely drives a robot so it can record demonstrations for a model to learn from.
Here's why this group matters now. A teleop lab exists to feed training data into a dedicated robotics model. If the company's position is that GPT-6 Astra can run robots directly, the reason that lab exists gets weaker. The people who run it know how to operate a data-collection pipeline at scale, and that's a skill any embodied-AI startup needs before it can train anything.
They also never appear on author lists. For robot learning engineer recruiting, the leads who ran collection pipelines can be more useful in the first year than another policy researcher, because a startup with no data has nothing to train on.
How do you pitch an engineer OpenAI is still paying?
Pitch status and ownership of the hardware, not job security. OpenAI's robotics salaries were reported at up to $310K, and one report puts base pay for some roles as high as $500,000. A seed or Series A startup won't win on cash, so it has to win on the things a model-first org can't give.
The levers that actually work:
- Their layer is the product. Brett Adcock's line, "In robotics, you've got to run the robot, see how it does," is the counter-argument to GPT-6 Astra running everything. At a hardware-first startup, the actuator engineer is central to the company, not a cost to minimize.
- The model bet may not win. Pete Florence of Generalist said it's "unclear that any existing model would be the one that dominates robotics in the long-run." An engineer who agrees has a reason to leave.
- Credit. Adcock said Figure's OpenAI partnership hurt hiring because candidates assumed Figure did the robotics and OpenAI made the models. The hardware team inside a model-first org runs into the same assumption. Offer them a place where nobody assumes the model did their work.
- Location. The Motors & Actuators roles are on site five days a week in SF. A hybrid schedule, or a team outside the Bay Area, is a lever OpenAI can't easily match.
- Timing. OpenAI alumni have founded robotics companies before. Covariant's founders, Abbeel, Chen, and Duan, had all worked at OpenAI, and Amazon did a reverse acquihire of them in 2024. The most capable applied engineers are the ones most likely to start something of their own, so reach them before they do.
The practical problem is speed. You're competing with a counteroffer in the hundreds of thousands, and finding hardware engineers who never publish is slow work. Handing the search to an agency adds a placement fee on top of a salary already set by OpenAI's numbers, and most early-stage founders can't absorb that for every hard-to-fill role.
FAQ
Did OpenAI lay off its robotics team?
No layoffs were reported. OpenAI moved robotics researchers under Chief Research Officer Mark Chen and applied robotics and hardware staff into Greg Brockman's Applied org. Michelle Fradin of OpenAI called the "disbanded" framing misleading. As of Nov 14, 2026, OpenAI was still posting robotics roles for control software, on-board computing, and motors and actuators, so anyone you contact is employed and should be approached with a poaching pitch.
How do I find OpenAI robotics hardware engineers if they don't publish papers?
Search where hardware engineers leave evidence: public code commits, actuator and stator patents, conference demo and booth staff, and past employers like Figure, Ford, and GM. Use OpenAI's own posting titles (stator process engineer, tooling/fixture engineer, firmware engineer) as search terms rather than "robotics" or "humanoid," since many of these engineers describe themselves by industry instead.
What salary do I need to compete with OpenAI for robotics engineers?
Reported OpenAI robotics salaries reach $310K, and one report cites base pay up to $500,000 for some roles. Most startups can't match that, so pitch what OpenAI can't easily offer: ownership of a hardware layer the company depends on, credit for the robotics work, and a hybrid or non-SF option for roles OpenAI requires on site five days a week.
The data behind this
Try it on the search you came here for
Stop building boolean strings. Just describe the person.
Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.
01Describe them
One plain sentence. Role, city, stack, stage, whatever matters to you.
02I read the web live
GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.
03You read the shortlist
Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.
- Staff backend engineers in NYC who shipped Rust in production
- Series A fintechs in SF under 50 people, growing headcount this year
- Maintainers of fast-growing Rust web frameworks on GitHub
- No boolean, no filters, no seat to buy. One box.
- Read at search time, so a profile updated yesterday counts today.
- Every step visible as it runs, every name with its reason.
500 free credits on sign-up. No card, no demo call. See real searches.