OpenAI, in addition to focusing on language models and applied artificial intelligence, is developing a much more discreet project related to robotics. The company has launched a secret laboratory in San Francisco with the aim of training robotic arms capable of performing everyday tasks.
The laboratory began operations in February last year and has grown significantly since then. According to reports, The space is now more than four times larger than when it began and operates 24 hours a day. Around a hundred data collectors and at least a dozen engineers specialized in robotics participate, working remotely controlling the robotic arms.
Unlike other companies that are betting directly on complete humanoid robots, OpenAI has opted for a more pragmatic strategy. The current focus is on low-cost robotic arms, trained to perform simple but essential actions in the home environment.. Tasks they learn include putting bread into a toaster or folding laundry, activities that require coordination, precision and understanding of the physical environment.
To generate the necessary data, human operators use a 3D printed controller called GELLO. This device directly translates human hand movements into robotic arm actionsallowing large volumes of real data to be collected.
Experts point out that the main challenge in robotics is not so much the software, but the availability of high-quality data from real interactions with the physical world. OpenAI appears to be addressing precisely that problem, laying the groundwork for future developments.
Although full humanoid robots are not an immediate priority, the company would have plans for a second robotics laboratory in California, evidence of its long-term interest in this area.
For the moment, the project remains out of the media spotlight, but it makes it clear that OpenAI is preparing the ground for a future in which artificial intelligence not only understands language, but can also interact with the physical world autonomously.