OpenAI Dissolves Independent Robotics Team, Researchers Merged into Core Research Team
According to a Fortune report, OpenAI has dissolved its independent robotics research team, merging the researchers into the core research team reporting to Chief Research Officer Mark Chen, while the applied robotics and hardware teams have been placed under the Applied division led by Greg Brockman; OpenAI has confirmed this adjustment. OpenAI stated that robotics remains central to its long-term vision and that this adjustment is not a withdrawal from the robotics field. The report states that this change is related to the judgment that general frontier models (including GPT-6 Astra) can already be used directly for robotics tasks; in an NVIDIA simulated robotic manipulation benchmark, the agent driven by Astra achieved a success rate of 74.8%, while the best publicly reported result for a specialized robotics model on that leaderboard is 51.0%.
Why read this
AI score 66
OpenAI has merged its robotics team into its general model system; the original article sorts out the changes in technical direction and industry disagreements behind this organizational adjustment.
Topics
Body · AI translation
OpenAI disbands its independent robotics team
Machine Heart Download the app for exclusive early access November 14, 2026, 13:38:01 From Beijing
Machine Heart Editorial Team
OpenAI's robotics team has suddenly been disbanded.
According to Fortune magazine, OpenAI has disbanded its independent robotics research team and reassigned the relevant researchers to other departments within the company.
The former robotics research team has been merged into the company's core research team, reporting to Chief Research Officer Mark Chen; the applied robotics and hardware teams have been placed under the larger Applied division, led by OpenAI President Greg Brockman.
This adjustment has been confirmed by OpenAI.
This week, at the CoRL robot learning conference held in Austin, USA, OpenAI had originally booked a booth; the conference app and the venue floor plan both listed OpenAI's name, but booth 208 ultimately stood empty, without even the company's logo appearing.
OpenAI emphasized that robotics remains a core part of the company's long-term vision, that this organizational adjustment is not a withdrawal from the robotics field, and that the company remains committed to bringing general-purpose robots into the real world.
Why did OpenAI disband its independent robotics team at this point in time?
This may be related to an important technical judgment. General-purpose large models have begun to demonstrate the ability to directly control robots, and OpenAI is trying to have general-purpose models take on more robotic intelligence tasks.
Why is OpenAI adjusting its robotics team again?
OpenAI's exploration of robotics has gone through several rounds of on-and-off efforts.
The company established a robotics team early on, but disbanded it in 2020. Co-founder Wojciech Zaremba later explained that at the time the robotics field faced the problem of limited training data, and that achieving AGI through other technical paths might be more efficient.
Subsequently, OpenAI shifted its R&D focus to large language models, training models on massive amounts of human text, and this path ultimately gave rise to ChatGPT, launched at the end of 2022.
By 2024, OpenAI once again formed an independent robotics team and announced a partnership with humanoid robot startup Figure to jointly develop AI models for humanoid robots. However, the partnership did not last long. In February 2025, Figure announced the end of the collaboration and turned to developing robot AI independently.
OpenAI then also began developing its own humanoid robots. At the end of May 2026, Sam Altman publicly posted about large-scale hiring, officially named the effort OpenAI Robotics, and set 2027 as the target time for its first public demonstration.
Looking at this development history, OpenAI's investment in robotics has not been interrupted; the changes have mainly occurred in technical approach and organizational structure.
According to Fortune, at the beginning of this year OpenAI's research thinking was still to seek breakthroughs in robotics technology from existing achievements such as the video generation model Sora.
To generate realistic videos, Sora needs to learn during training about object motion, spatial relationships, and some of the laws of the physical world, and these capabilities have the potential to transfer to the robotics field.
But afterward, OpenAI found that general-purpose frontier models, including GPT-6 Astra, were already able to be applied directly to robotics tasks.
This discovery changed the company's R&D thinking. OpenAI decided to merge robotics researchers into the general-purpose model R&D system, so that robotics research would be more closely integrated with advances in foundation model capabilities.
This reflects a technical approach that is gradually gaining recognition in the AI industry. In the past, robots typically required specially trained models to understand the environment, plan actions, and control actuators, and different robot hardware, tasks, and application scenarios often required large amounts of data for adaptation. Today, general-purpose models are beginning to show the potential to cross these task boundaries, and foundation models trained for language, code, and vision also have the opportunity to be used directly to control devices in the physical world.
Can general-purpose large models take over the robot brain?
This approach has already produced some surprising experimental results.
Recently, researchers tried connecting GPT-6 Astra to a robot, providing the model with camera images, robot-related information, and an interface capable of controlling the robot's actions.
In an NVIDIA study, an agent powered by Astra achieved a 74.8% success rate on a simulated robot manipulation benchmark. By comparison, the best-performing specialized robotics model on that leaderboard has a publicly reported score of 51.0%.
This experiment was completed in a digital simulation environment, and the model controlled a simulated robot, not a real one. Nevertheless, these results still demonstrate the potential of general-purpose models on robotics tasks.
Guanya Shi, an assistant professor at Carnegie Mellon University's Robotics Institute, believes that OpenAI's adjustment this time sends an important signal. OpenAI may no longer view robotics as a special problem that needs to be solved independently, but rather as a downstream application of general intelligence. This is similar to today's AI coding, legal, and financial tools: they use the same underlying general-purpose model, and then adapt to specific tasks through different tools and interfaces.
Advances in general-purpose model capabilities also give robots an opportunity to reduce their dependence on task-specific training.
For example, when the position of a bottle changes, or the ambient lighting changes somewhat, the model can still adapt to a certain extent. They can also reason about the order in which tasks should be executed, determine whether a certain step has gone wrong, and adjust subsequent actions accordingly.
However, there are still many limitations before general-purpose models can fully take over robot control.
The first is inference efficiency. Although large language models are powerful, they run relatively slowly and usually still need to rely on cloud computing. For robots that require real-time responses, network latency and connection stability are real problems.
Guanya Shi believes that small, specialized models that can run on a robot's local computing devices still have important value, as they can complete some tasks without relying on a continuous network connection.
The second is data and real-world experience.
Russ Tedrake, an MIT professor and co-founder and CEO of Walden Robotics, points out that robotics companies have an advantage that AI labs find difficult to replicate: their robots have already entered real scenarios such as factories and are continuously accumulating practical work experience. These deployment data can help robots adapt to specific workplaces and specific tasks. The stronger the foundation model capabilities, the more efficiently robots may be able to learn from these data.
In other words, general-purpose foundation models and real-world scenario data still need to develop in tandem: the former provides stronger understanding and reasoning capabilities, while the latter helps robots adapt to specific environments and tasks.
Pete Florence, co-founder of Generalist, a startup focused on general-purpose robot AI models, also believes that it is still impossible to determine which technical route will ultimately dominate the robotics field. Intelligence may have different starting points for growth, and existing models are not yet sufficient to prove who can become the long-term winner.
Is the robotics industry's technical route about to change?
OpenAI's adjustment this time may also have an impact on the entire robotics industry.
Over the past few years, the robotics field has attracted billions of dollars in investment, and many startups have bet on a similar route: building a dedicated AI brain for robots, collecting robot data at scale, and training specialized models so that robots gradually acquire general manipulation capabilities.
OpenAI is now trying another possibility. General-purpose models first acquire sufficiently strong understanding, reasoning, and spatial capabilities, and then connect to physical devices through interfaces to directly take on the intelligent control tasks of robots.
If this route continues to make progress, the R&D systems built around specialized models in the past will need to be reassessed. However, the value of robotics companies will not disappear because of this. Real-world deployment, hardware design, control systems, and scenario data remain important foundations for robots to move toward large-scale application.
OpenAI's adjustment this time also comes against the backdrop of the company facing pressure to control costs.
Last year, Altman asked the company to abandon some "side quests" that deviated from its core goals and to concentrate resources on more important products and R&D directions. Since then, OpenAI has successively adjusted some projects in order to reallocate computing resources.
According to reports, OpenAI's operating loss in the most recent quarter reached $10.9 billion. At the same time, the company is seeking a new round of financing and preparing for a possible IPO in early 2027.
Against this backdrop, reintegrating researchers scattered across different teams and concentrating investment in general-purpose model R&D is also consistent with the company's approach of improving resource utilization efficiency.
OpenAI's judgment about robotics technology is changing. In the future, will robots need a dedicated brain trained independently, or can they directly call on ever-evolving general intelligence?
"Special statement: The content of the above works (including videos, pictures, or audio if any) is uploaded and published by users of "Dafeng Hao," a self-media platform under Phoenix Net. This platform only provides information storage space services.
Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."