OpenAI/unsplash
Startup “Figure” demonstrated the first results of joint work with the OpenAI company on expanding the capabilities of humanoid robots. Their rapid progress is admirable.
In the published video, the robot engages in a real-time dialogue with a human, answering questions and following commands. This is written by one of the world's largest independent scientific and technical publications, New Atlas.
According to the founder of the startup, Brett Adcock, the cameras built into Figure 01 transmit data to a large visual-language AI model trained by OpenAI. While Figure's proprietary neural networks also capture images of the environment at 10 Hz through the robot's cameras.
OpenAI algorithms are responsible for the robot's ability to understand human language, while Figure's neural network transforms the flow of information into fast, low-level and nimble actions of the robot. The head of the project stated that during the demonstration the robot was not controlled remotely and the video was shown at real speed.
Our goal is to teach the world model of AI to control humanoid robots at the level of billions of units, the head added startup.
At this rate of development, there is not much time left for the project to be completed.