OpenAI

• ChatGPT and Whisper with Ameca Desktop

  Reference Design - Controlled By - Engineered Arts documents OpenAI ChatGPT and Whisper integration in Ameca Desktop.

  Source

• OpenAI trains robotic hand to solve a Rubik's Cube with Automatic Domain Randomization

  2019-10-15 - Research Publication - OpenAI trained neural networks to solve a Rubik's Cube with a human-like robotic hand using reinforcement learning in simulation. The project introduced Automatic Domain Randomization to create progressively harder simulated environments for sim-to-real transfer. The system solved simpler scrambles 60 percent of the time and maximal-difficulty scrambles 20 percent of the time, making it a durable dexterous-manipulation milestone.

  Source
• OpenAI trains Dactyl robotic hand system for dexterous manipulation

  2018-07-30 - Research Publication - OpenAI introduced Dactyl, a system that trains a Shadow Dexterous Hand entirely in simulation and transfers the policy to real hardware. The work used the same general-purpose reinforcement learning system behind OpenAI Five to manipulate objects into target orientations. The event remains a landmark robotics result because it showed sim-trained dexterous manipulation could adapt to real-world physics.

  Source

Events

• 2026-05-10 - Capability - OpenAI robotics hiring points to scaled data-acquisition and workcell operations

  OpenAI listed a Robotics Technical Program Manager role focused on robotic data acquisition in San Francisco. The posting describes DAQ stations, robotic platforms, sensing systems, operator tooling, and data pipelines for model training and evaluation. The hiring signal supports the view that OpenAI is building operational infrastructure for robotics data collection across development and evaluation environments.

  Source
• 2026-02-05 - Demonstration - Ginkgo and OpenAI run 36,000 autonomous lab experiments

  The system linked GPT-5 with Ginkgo's Reconfigurable Automation Carts and Catalyst software to design, execute and analyse experiments with minimal human involvement. Across six iterative cycles, the autonomous laboratory processed more than 580 384-well plates and generated nearly 150,000 experimental data points. The resulting reaction mix produced the benchmark protein sfGFP at $422 per gram in component costs, compared with a previously reported $698 per gram.

  Source
• 2026-01-15 - Market Signal - OpenAI launches U.S. AI supply-chain RFP covering advanced robotics inputs

  OpenAI launched a U.S. manufacturing RFP to identify domestic suppliers for broader AI infrastructure. The RFP explicitly includes critical inputs for advanced robotics, including gearboxes, motors, and power electronics. The row is a supply-chain signal rather than a product launch because it reflects demand formation around robotics-enabling components.

  Source
• 2025-12-16 - Partnership - Robot on Rails, OpenAI and Red Queen Bio test AI-directed wet-lab research

  The collaboration connected GPT-5 protocol design with Robot on Rails hardware executing molecular-cloning experiments on a conventional lab bench. The AI-generated protocol improved the target metric 79-fold, while robotic execution produced a 2.13-fold experimental improvement and reached 89% of the relative performance achieved by a human scientist. Absolute colony counts remained about ten times lower than the human baseline, providing important context on the gap between relative optimisation and end-to-end laboratory performance.

  Source
• 2024-12-10 - Award - Camfer wins OpenAI o1 hackathon with automated airfoil design

  The demonstration generated five airfoil profiles for 50 mph operation with a requested minimum lift-to-drag ratio of 15 at a five-degree angle of attack. The system iterated between spline generation, simulated performance testing and design updates. Camfer also presented the demonstration at OpenAI DevDay 2024.

  Source
• 2019-10-15 - Research Publication - OpenAI trains robotic hand to solve a Rubik's Cube with Automatic Domain Randomization

  OpenAI trained neural networks to solve a Rubik's Cube with a human-like robotic hand using reinforcement learning in simulation. The project introduced Automatic Domain Randomization to create progressively harder simulated environments for sim-to-real transfer. The system solved simpler scrambles 60 percent of the time and maximal-difficulty scrambles 20 percent of the time, making it a durable dexterous-manipulation milestone.

  Source
• 2018-07-30 - Research Publication - OpenAI trains Dactyl robotic hand system for dexterous manipulation

  OpenAI introduced Dactyl, a system that trains a Shadow Dexterous Hand entirely in simulation and transfers the policy to real hardware. The work used the same general-purpose reinforcement learning system behind OpenAI Five to manipulate objects into target orientations. The event remains a landmark robotics result because it showed sim-trained dexterous manipulation could adapt to real-world physics.

  Source

Profile notes

• OpenAI left in-house robotics over a data bottleneck, then rebuilt the capability around broader model and hardware work

  reviewed 2026-07-20 - OpenAI disbanded its original robotics group after concluding that scarce physical-interaction data limited progress, shifting attention to more scalable AI research. It restarted a robotics team in 2024 and now advertises work across the model stack and multiple hardware form factors, making the current effort a reboot under changed technical assumptions rather than uninterrupted continuation of the Rubik's Cube hand programme.

  Source
• Gymnasium is the maintained community successor to OpenAI Gym, not a new OpenAI release

  reviewed 2026-07-20 - Maintenance of OpenAI Gym declined until OpenAI handed the repository to an outside team and stopped planning further development. That team created the Farama Foundation and moved future maintenance to the Gymnasium fork, which preserves the core interface while introducing API changes; active Gymnasium development is therefore community continuity beyond OpenAI rather than evidence that OpenAI still maintains Gym.

  Source

Relationships

• Figure AI

  Technology Partner - Figure and OpenAI agreed to develop next-generation AI models for humanoid robots.

• Robot on Rails

  AI Directed Wet Lab Research - Robot on Rails worked with OpenAI and Red Queen Bio on an AI-directed wet-lab molecular-cloning system.

• Ginkgo Bioworks

  Gpt 5 Autonomous Laboratory Research Partner - Ginkgo Bioworks and OpenAI connected GPT-5 to Ginkgo's autonomous laboratory for closed-loop biological experimentation.