OpenAI Disbands Independent Robotics Team, Researchers Merged into Core Research Team
According to media reports, OpenAI has disbanded its independent robotics team. The former robotics research team has been merged into the core research team and will report to Chief Research Officer Mark Chen, while the applied robotics team and hardware team have been placed under the applied division led by Greg Brockman. An OpenAI spokesperson confirmed the news, stating that this does not mean the company is exiting the robotics field. This is the second time OpenAI has disbanded its robotics team; it previously rebuilt the team in 2024 and partnered with Figure, with the two sides ending their collaboration in February 2025. The company had previously disclosed plans to publicly demonstrate its self-developed humanoid robot for the first time in 2027. The report links this adjustment to the ability of general-purpose models such as GPT-6 Astra, released in September, to directly take on the work of robot policy models.
Why read this
AI score 63
OpenAI disbands its robotics team for the second time; reports contrast this organizational adjustment with progress in general-purpose models directly controlling robots.
Topics
Body · AI translation
Main text
According to media reports, OpenAI has disbanded its independent robotics team and reassigned the team's researchers to other departments.
An OpenAI spokesperson confirmed the news and stated that the move does not mean the company is exiting the robotics field, and that robotics technology remains a core component of OpenAI's long-term vision.
OpenAI's original robotics research team has been merged into the company's core research team, reporting to Chief Research Officer Mark Chen; the applied robotics team and the hardware team have been reassigned to the company's larger applications division, which is led by OpenAI President Greg Brockman.
This is the second time OpenAI has disbanded its robotics team.
In its early days, OpenAI had a robotics team, but it gradually disbanded the team in 2020. Co-founder Wojciech Zaremba later explained that at the time, available data in the robotics field was limited, and achieving AGI through other technical paths was more efficient.
Until 2024, OpenAI rebuilt an independent robotics team and announced a partnership with the startup Figure to jointly develop AI models for humanoid robots. The two sides ended their collaboration in February 2025. At the time, Figure said it would develop the robot's AI system on its own.
OpenAI previously revealed that it is independently developing a humanoid robot and plans to hold its first public demonstration in 2027.
Do robots no longer need a dedicated "brain"?
The variable appeared with the release of GPT-6 Astra in September.
This model demonstrated the ability of a general-purpose multimodal foundation model to directly take on part of the work of a robot policy model, causing a stir in the industry. Wang Xingxing, founder of Unitree Robotics, also revealed that Unitree has deployed GPT-6 on robots for task testing with good results, and expects that after the release of GPT-7 next year, it may further break through the current bottlenecks of embodied intelligence.
This week's release of Anthropic's new-generation flagship model Fable 5.2 further confirms this trend. Also a general-purpose large model, Fable 5.2 has shown greatly improved ability in controlling physical robots, even including tasks it was never explicitly trained on.
Previously, the mainstream view held that robots needed more vertical large models designed specifically for embodied intelligence to effectively solve spatial understanding, action planning and control. Now, as the general-purpose models of tech giants continue to make breakthroughs, another consensus is beginning to form in the AI industry: general-purpose models developed for language, code and vision tasks can also be applied to the field of embodied intelligence.
Guanya Shi, an assistant professor at Carnegie Mellon University's Robotics Institute, said in an interview with the media that he believes OpenAI's disbanding of its robotics team can be interpreted as a signal: OpenAI believes that robotics technology is "no longer a special problem."
He believes that OpenAI likely views robotics as a problem that can be solved through downstream applications of general intelligence. This is similar to the current situation: the same set of underlying frontier AI models can already support specialized programming tools as well as specialized tools in the legal and financial fields.
OpenAI's own response is consistent with this.
The company said that at the beginning of this year, researchers believed that the best way to advance robotics technology was to continue building on the company's existing AI model research, such as the video generation model Sora. To generate realistic videos, Sora must have some implicit understanding of how the physical world operates.
But since then, the company's general-purpose frontier models (including GPT-6 Astra) have demonstrated the ability to be applied directly to the robotics field. It was precisely this discovery that prompted OpenAI to disband its independent robotics team and reassign the relevant researchers to teams responsible for developing more general-purpose AI models.
However, this does not mean robotics startups have no way forward. Russ Tedrake, a professor at MIT and co-founder and CEO of Walden Robotics, said that robotics companies still have an advantage that AI labs do not: their robots have already accumulated practical work experience in real-world scenarios such as factory floors. This deployment experience, as well as the data collected by robots during actual operation, helps robots adapt to specific workplaces and tasks. Tedrake said that the more powerful the foundation models become, the more valuable this scenario-specific data will be, because robots can learn from it faster.
(Source: Star Market Daily) Middle of the main text Embedded advertisement