Inspenet, December 13, 2023.
A group of researchers has linked a new humanoid robot known as Alter3 , which has been developed by a team at the University of Tokyo, with GPT-4 (OpenAI’s artificial intelligence model).
Thanks to this connection, the humanoid now has the ability to assume various poses, such as those designed for selfies or even simulate being a ghost, without the need for these to be previously programmed in its database.
Until now, controlling low-level aspects of robots, such as movement, has been a challenge for large language models, as it involves hardware aspects that are not included in the training data. To overcome this obstacle, the Japanese team has devised a way to translate the expression of human movement into a code understandable by the humanoid .
This enabled Alter3 to generate spontaneous movements even though he is not currently trained to perform complex movements, since his mobility is limited to the upper part of his body and his lower section is fixed to a support. This approach means the robot can generate sequences of actions over time without requiring developers to program each specific body movement. It is able to learn to move in a manner similar to a small child, progressively moving from shuffling to crawling and finally walking.
Users also have the ability to adjust posture or help the android discern between various postures. If a human gives instructions to Alter3, such as “take a selfie with your iPhone,” the robot will execute a series of requests to OpenAI’s GPT-4 model, including asking the AI what taking a selfie entails. GPT-4 then translates this into Python code that the robot can understand and use to perform the necessary movements.
It should be noted that the group responsible for Alter3 and the novel text-to-motion translation strategy based on GPT-4 intends to share their discoveries so that they can be used in other humanoid robots. According to them, this represents a notable step forward in the progress of robotics.
“Surprisingly, Alter3 can imitate ghosts, animals, and human-like expressions of emotions. Its response to the content of a conversation through facial expressions and gestures represents a significant advance in humanoid robotics, easily adaptable to other androids with minimal modifications.” they wrote in the research article.
Don’t miss any of our posts and follow us on social media!