AI News

OpenAI Dissolves Standalone Robotics Team, Researchers Merged into Core Research Division

Phoenix TechnologyUpdated Nov 14 17:51

AI summary

OpenAI confirmed to Fortune that it has dissolved its standalone robotics research team. The former robotics researchers have been moved into the core research division led by Chief Research Officer Mark Chen, while robotics applications and hardware staff have been moved into the Applied division headed by President Greg Brockman. The company said robotics remains part of its long-term technology vision and that the adjustment is intended to accelerate R&D progress, because general-purpose models such as GPT-6 Astra have already demonstrated the ability to directly control robots. Previously, OpenAI had built a robotics data collection lab of about 100 people in San Francisco, and in September some robotics positions offered base annual salaries of up to $500,000.

Why read this

AI score 72

This article reviews the investment and restructuring details in the year before and after OpenAI dissolved its robotics team, and compares them with the specialized model approaches of Google and Nvidia, presenting the divergence over general-purpose models taking over robot intelligence.

Topics

Tags

Body · AI translation

Just after offering annual salaries of 3.3 million yuan to poach talent, OpenAI suddenly dissolved its robotics team—is the industry about to change! Phoenix Tech Download the app for exclusive early access November 14, 2026, 17:14:24 From Beijing

On November 13, at the Conference on Robot Learning (CoRL) held in Austin, USA, booth 208 that OpenAI had reserved stood empty.

The conference booth map still marked OpenAI, but no company logo or product display was seen.

Subsequently, OpenAI confirmed to Fortune that it had dissolved its independent robotics research team.

The former robotics researchers moved into the core research department led by Chief Research Officer Mark Chen, while robotics applications and hardware staff moved into the Applied department headed by President Greg Brockman.

The company said robotics remains an important part of its long-term technology vision and that this adjustment is intended to accelerate R&D progress.

The news came somewhat suddenly.

In September this year, OpenAI was still recruiting robotics R&D staff on a large scale, with base annual salaries for some positions as high as $500,000. At the start of the year, the company's San Francisco robotics data collection lab already had about 100 workers, and a second lab was being planned.

Sam Altman had also publicly said he expected to be able to show an impressive robotics demo in 2027.

Investing in labs and hiring hardware engineers on one hand, while dissolving the independent research team on the other—how exactly does OpenAI plan to do robotics? And where will the R&D resources invested previously flow?

How much has OpenAI invested in embodied AI?

OpenAI has in fact already had one experience of dissolving a robotics team.

In 2019, the company demonstrated the Dactyl robotic hand solving a Rubik's cube. Researchers used large-scale simulation and reinforcement learning to enable the robot to master complex object manipulation capabilities.

Image | In 2019, OpenAI demonstrated research on controlling a robotic hand to solve a Rubik's cube through reinforcement learning

In 2020, OpenAI ended its first round of independent robotics research and shifted more effort to large-scale language models. That path later gave rise to ChatGPT.

In 2024, the company re-formed a robotics team and partnered with humanoid robotics company Figure to develop AI models. The partnership ended in February 2025; Figure shifted to developing its own robotics models, while OpenAI continued to expand internal R&D.

In January this year, Business Insider investigated OpenAI's robotics data collection lab in San Francisco. The lab began construction in February 2025 and has since expanded more than fourfold, with about 100 data collection workers operating in three shifts.

Workers use 3D-printed GELLO teleoperation controllers to operate Franka robotic arms to complete tasks such as folding clothes and putting bread into a toaster. The system records scenes and operation processes through cameras, accumulating data for training robotics models.

The lab's early tasks were fairly simple, including putting a rubber duck into a cup, and only later did the difficulty of operations gradually increase.

The report also mentioned a detail: although there was a humanoid-robot-like device in the lab, the main data collection work was still done with robotic arms.

OpenAI also planned to build a second lab in Richmond, California.

For embodied intelligence companies, data collection is a considerable expense. Language models can use the large amount of text and images already available on the internet, whereas robots need to learn physical operations such as grasping, contact, and movement. Completing this data collection requires equipment, personnel, maintenance, and on-site management.

At the time, OpenAI was building its own data production system, and the scale of hiring was gradually expanding.

In September this year, Business Insider counted that OpenAI's robotics-related job openings had increased from 11 in May to 27, covering directions such as machine learning, robot control, electronic engineering, actuators, firmware, and data collection, with base annual salaries for some positions reaching as high as $500,000.

As of November 14, a Phoenix Tech check of the official website found that OpenAI still publicly listed multiple robotics-related engineering positions.

Image | As of November 14, OpenAI's official website was still recruiting for embodied-related positions such as onboard robot computing, control software, motors and actuators

Previously, the company was also advancing supply chain preparations. In January this year, OpenAI released a call for U.S. manufacturing suppliers, listing robotics together with data centers and consumer electronics in its hardware supply chain layout.

The solicitation covered key components such as motors, gearboxes, reducers, and power electronics. According to the original document, OpenAI planned to select suppliers in 2027, but it has not disclosed the amount of the relevant procurement contracts.

In addition to engineering R&D, funding arrangements for the robotics business had also entered management discussions.

In May this year, The Wall Street Journal disclosed that Sam Altman had considered operating the robotics and consumer hardware businesses independently, allowing the relevant companies to raise funds separately from external investors.

The plan was ultimately not implemented. One reason was that even if the business were spun off, the relevant assets and profit and loss might still need to be included in OpenAI Group's consolidated financial statements.

These records show that over the past year OpenAI's investment in robotics has already involved data collection, model research, hardware engineering, and supply chain preparation.

After confirming the reorganization on November 13, OpenAI has not yet announced specific allocations for budget, personnel, or compute, nor has it said whether the data collection lab and prototype projects will be adjusted.

OpenAI explained the technical background of this adjustment to Fortune. At the beginning of this year, the company's researchers originally believed that using video generation models such as Sora to learn the laws of the physical world was an important path toward advancing robot intelligence.

But general-purpose models such as GPT-6 Astra subsequently demonstrated the ability to directly control robots. OpenAI decided to merge the relevant researchers into the general foundation model team, allowing robotics research to leverage the latest advances in existing models.

Two speeches by Unitree founder Wang Xingxing over the past month also align with this trend.

On October 12, during a roundtable discussion at the Shanghai Industrial Expo, Wang Xingxing said that after the release of GPT-6, Unitree had already integrated it into various robots to perform tasks, and the actual results exceeded previous expectations.

Image | On October 22, 2026, Wang Xingxing shared progress on general large models controlling robots at the World Physical AI Conference. Image source: The Paper

Wang Xingxing believes that AI models show noticeable progress almost every one or two months. If GPT-7 continues to achieve breakthroughs, the technical tipping point for embodied intelligence may arrive in the next one to two years.

On October 22, at the World Physical AI Conference, he again demonstrated an experiment in which a Unitree robot used GPT-6 to complete a painting. In front of the robot were a reference painting, white paper, paint, and clean water. It picked up a brush to paint, switched colors during the process, and even washed the brush by itself. Unitree also demonstrated tasks such as the robot playing chess.

Wang Xingxing repeatedly emphasized that the team did not conduct specialized training for these tasks, nor did it write complex motion programs.

The robot itself still requires motion control and hardware systems, but general models are already able to organize some operational steps based on visual input and task requirements.

Previously, many robotic tasks required developers to collect specialized motion demonstrations and then train corresponding manipulation policies. Now, general models can handle part of the visual understanding, spatial reasoning, and task planning work.

Anthropic's recently released Claude Fable 5.2 also provides experimental data.

On November 9, Anthropic demonstrated Claude controlling a robotic arm to solve a Rubik's cube, unscrew a bottle cap, and paint.

In the RoboDojo Sim simulation benchmark, tests run internally by Anthropic showed that Fable 5.2 achieved an average success rate of 64.9%, compared with 27.6% for the previous generation Fable 5.1, while GPT-6 Astra scored 29.8% under the same testing framework.

Image | Anthropic's published RoboDojo Sim simulation evaluation, in which Claude Fable 5.2 achieved a score of 64.9%. Image source: Anthropic

NVIDIA's Astranomy research, published in October, directly used GPT-6 Astra to control simulated robots.

In the RoboLab-120 test, the direct control approach achieved a task success rate of 66.3%. After researchers added mechanisms such as continuously recording operation history and checking execution results, the agent control approach reached 74.8%, higher than the previous best specialized robot model score of 51.0% on that benchmark.

Although these experiments used different testing environments, they all demonstrated progress in general models' physical manipulation capabilities.

Guanya Shi, a robotics researcher at Carnegie Mellon University, said in an interview with Fortune that OpenAI may be viewing robotics as a downstream application of general intelligence.

For OpenAI, directly using general models means that some robotics intelligence research can share the training results of core models. The return on investment of independently developing a set of robot foundation models therefore needs to be reassessed.

However, the NVIDIA experiment also exposed another problem: the computational cost of general models is quite high.

In the Astranomy research, the direct control approach with GPT-6 Astra took an average of 18.56 seconds per inference.

Although the agent approach improved the success rate, each successful task episode consumed an average of about 30.9 million tokens, corresponding to an estimated API cost of $36.40.

These costs come from simulation tests and cannot be directly treated as the operating costs of real robots. But in actual factories, the longer a robot takes to perform a task, the higher the cost per unit of output may be.

Even more troublesome is latency. Simulation environments can pause during model inference, but conveyor belts, equipment, and workpieces in real factories will not stop and wait for more than ten seconds.

Robots still need systems that respond faster and can directly control the body's movements. This is also one of the reasons other tech companies continue to train specialized robot models.

Google, NVIDIA, and Unitree are still training specialized models

On November 10, Google DeepMind announced an experiment in which the Apollo 2 humanoid robot worked in a Mercedes-Benz factory.

The robot grabbed metal brackets from bins and placed them into designated containers according to the production sequence. The test locations included Mercedes-Benz factories in Sindelfingen, Germany, and Kecskemét, Hungary. Google announced that the success rate for this parts sorting task reached 96%.

Image | The Apollo 2 humanoid robot conducting a parts sorting test at a Mercedes-Benz automobile factory. Image source: Google DeepMind

During operation, two sets of models each performed their own functions.

Gemini Robotics ER is responsible for understanding the production process and judging task progress; Gemini Robotics 2.5, trained on robot actions, is responsible for the actual operation of the arms and dexterous hands. Google also uses Gemini 4 Argon to help build simulation environments and analyze failure cases.

Apptronik CEO Jeff Cardenas said that Apollo 2 showed marked improvement after several weeks of post-training, and that performance degradation was relatively small after migrating to the Mercedes-Benz production line.

This is a proof of concept in a limited scenario, and long-term failure rates and production costs have not yet been disclosed. But Google has demonstrated how a general reasoning model and a specialized action model can work together to complete factory tasks.

NVIDIA's GR00T 2, released in November, focuses on the training cost after robots enter new scenarios.

Image | GR00T 2 reads recent videos, instructions, and robot states, performs denoising on future video and action segments, executes that action segment, and then repeats this process

In 50 tasks that had not previously been trained for the target robot, GR00T 2 achieved an average zero-shot success rate of 40%. The researchers added 5 demonstration operations and less than 3 minutes of robot data for each task, then used a single RTX 5090 to fine-tune for about 30 minutes, raising the success rate to 95%.

When robots arrive at a new factory, they often need to re-adapt to item placement, grasping order, and specific equipment. Being able to complete task adaptation with only a small number of demonstrations offers an opportunity to reduce investment in data collection, on-site debugging, and engineers stationed on site.

But robots still have to solve physical errors in precision operations.

On November 5, Wang Xingxing mentioned at the Hongqiao Forum of the China International Import Expo that current top multimodal models already have strong understanding and planning capabilities, but tasks such as component insertion are still prone to getting stuck in the last few millimeters.

Picking up a box can tolerate a certain positional deviation, while inserting a connector must handle angle, friction, and contact force. A slight deviation in motion may damage the workpiece.

NVIDIA conducted real-machine reinforcement learning for these tasks. In experiments it published, trained robots achieved a 94% success rate in cable insertion tasks and 96% in charger insertion.

Unitree and AgiBot are also continuing to train their own robot models. Unitree's publicly disclosed UnifoLM-WLA-1.0 uses about 2,500 hours of real-machine data, while AgiBot continues to invest in robot operation data and post-training through GO-3 Preview and Genie Evolver.

In a typical precision assembly task, a general model can understand the operation requirements and arrange the execution order, while a specialized action model needs to learn how different robotic arms approach parts, adjust positions, and handle contact feedback. To operate continuously, the underlying control system also needs to handle exceptions in a timely manner.

This division of labor does not require every company to train a set of foundation models for understanding and planning from scratch. Robot companies can use external models and then train action capabilities for their own bodies, customer tasks, and on-site data.

For OpenAI, incorporating robotics research into the GPT team is precisely a choice to solve more robot intelligence problems within the general model system. The data collection and hardware R&D capabilities already built can then be used to test the effects of these models on real robots.

Source

Phoenix Technology