OpenAI Dissolves Independent Robotics Research Team, Staff Merged into Core Research and Applications Departments
OpenAI has dissolved its independent robotics research team: robotics researchers have been merged into the core research team led by Chief Research Officer Mark Chen, while the robotics applications and hardware teams have been transferred to the applications business department headed by President Greg Brockman. OpenAI stated that general-purpose models, including GPT-6 Astra, can already directly perform some robot control tasks, so it no longer relies entirely on independent robotics-specific foundation models. The article mentions that a NVIDIA study showed that an agent system based on GPT-6 Astra achieved a task success rate of 74.8% in the RoboLab-120 simulation test, higher than the 51.0% of the best specialized robotics model on that leaderboard, but this result depends on an additional agent runtime environment and was achieved only under simulation conditions. This reorganization comes as OpenAI faces pressure to control costs, and the company still plans to develop humanoid robots independently.
Why read this
AI score 70
OpenAI has merged its robotics team into its core research and applications departments, allowing readers to observe the trade-offs between general-purpose models and specialized robotics approaches.
Topics
Body · AI translation
Under the arrangement, the robotics researchers will be merged into the core research team led by Chief Research Officer Mark Chen, while the robotics applications and hardware teams will move to the applications business division headed by President Greg Brockman.
This adjustment reflects a change in OpenAI's assessment of its robotics technology roadmap. The company said that general-purpose AI models, including GPT-6 Astra, have demonstrated the ability to directly perform some robot control tasks, so it no longer needs to rely entirely on independently developed robotics-specific foundation models.
A recent study published by Nvidia showed that in the RoboLab-120 simulated robot manipulation test, an agent system based on GPT-6 Astra achieved a task success rate of 74.8%, higher than the 51.0% of the best-performing specialized robotics model on that test's leaderboard. However, this result relies on an additional agent runtime environment and was achieved only under simulation conditions. General-purpose models still have clear limitations in inference speed and operating cost, and cannot yet directly replace specialized robot control systems in real-world scenarios.
The restructuring also comes as OpenAI faces pressure to control costs. The company's operating loss in the most recent quarter is said to have reached $10.9 billion, and it is currently pursuing a new round of financing and preparing for a possible initial public offering in early 2027. Despite continuing to scale back some non-core businesses, OpenAI still plans to develop humanoid robots on its own, and CEO Altman previously proposed aiming for a first public demonstration in 2027.