OpenAI is reportedly returning to robotics, a field it stepped back from in 2020. The move centers on a rebuilt team, early hiring, and a strategy that appears designed to put OpenAI models inside robots made by other companies.
The reported restart is still at an early stage. Several sources told Forbes that OpenAI is hiring research engineers to rebuild its robotics division, and the new team has only been in place for about two months.
A robotics division returns after 2020
OpenAI shut down its robotics department in 2020. Now, according to the source report, the company is bringing that work back with the goal of speeding up the release of AI-powered robots.
The hiring signal is direct. A recent job posting said a new hire would be "one of the first members of the team." That wording suggests OpenAI is still assembling the foundation of the group rather than expanding a mature operation.
The reported focus is also narrower than building an entire robot business from scratch. OpenAI's goal is to collaborate with robotics companies, not compete with them. In practice, that means creating technology that robot makers can incorporate into their own products.
Why Figure AI matters to the strategy
The clearest example in the source is OpenAI's work with Figure AI, a robotics startup. Figure's robot uses an OpenAI AI multimodal model that can process several kinds of input, including audio, vision, and language.
The source notes that the model might be the new GPT-4o. The important point is not just the possible model name, but the role of multimodal AI in robotics. A robot that can process audio, vision, and language has a different software foundation than a machine that only follows narrow programmed instructions.
For OpenAI, that kind of collaboration allows its technology to reach physical systems without requiring the company to own every part of the robotics stack. For robot makers, the appeal is access to AI models that can interpret different forms of information in one system.
Peter Welinder, VP of Product and Partnerships at OpenAI, described the direction in a statement to PopSci after Figure got its latest round of money in February: "We've always planned to come back to robotics and we see a path with Figure to explore what humanoid robots can achieve when powered by highly capable multimodal models,"
OpenAI's first robotics push came too early
OpenAI's return to robotics did not come without hesitation. OpenAI's CEO, Sam Altman, discussed getting back into robotics research in a podcast with Microsoft founder Bill Gates in January. According to Altman, the company's first attempt was premature.
The earlier shutdown had practical reasons. OpenAI stopped its own robotics research to focus on making its AI models bigger and better. At the time, co-founder Wojciech Zaremba said the company also stopped because it did not have enough data to train capable robot models.
That context helps explain why the new approach matters. If OpenAI is rebuilding robotics through partnerships and model technology, it may be trying to avoid some of the hard limits of its earlier effort. The source does not say whether those limits have been fully solved, but it does show that OpenAI now sees a path back into the field.
The hardware question is still open
One of the biggest unanswered questions is whether OpenAI will manufacture its own robot parts. The source says that is not yet clear, and notes that OpenAI has struggled with that in the past.
That uncertainty matters because robotics is not only a software challenge. Even powerful AI models need physical systems that can move, sense, and operate reliably. The source article does not provide details on whether OpenAI will build those systems itself, rely on partners, or use some mix of both.
What is clear is the stated collaboration-first posture. OpenAI wants to create technology that robot makers can use in their own products. That makes robotics companies central to the effort, at least based on what has been reported so far.
A field OpenAI never fully left behind
Robotics was important to OpenAI in its early years. In 2019, the research team wrote a paper about how a robotic arm with neural networks learned to solve a Rubik's cube on its own.
That example shows why robotics has long been relevant to OpenAI's broader ambitions. Robots bring AI out of purely digital environments and into physical tasks, where vision, language, audio, motion, and decision-making can all matter at once.
The reported rebuilding of OpenAI's robotics division is therefore not just a staffing update. It is a sign that the company is again looking at how AI models can work in machines that interact with the real world. For now, the available facts point to a young team, early hiring, collaboration with robotics companies, and a bet on multimodal models as the bridge between AI research and humanoid robots.