Researchers at the University of Tokyo have developed an advanced robot friend named Alter3 that possesses natural language abilities and human-like behaviors, thanks to the power of GPT-4, the latest and most advanced large language model in the world. Alter3 is a humanoid robot that was initially introduced in 2016 to explore the concept of life in artificial systems. It has a realistic appearance and can move its upper body, head, and facial muscles, with 43 axes controlled by air actuators. Alter3 also has cameras in its eyes, enabling it to see and interact with humans and its environment.
What sets Alter3 apart is its ability to utilize GPT-4, a deep learning model, to control its movements and behaviors. Instead of meticulously programming every action, researchers can now give verbal instructions to Alter3, and GPT-4 generates the corresponding Python code that operates the Android engine. For instance, to make Alter3 take a selfie, researchers can simply give it instructions like creating a joyful smile, widening its eyes, adopting a dynamic posture, and simulating holding a phone. GPT-4 generates the code that makes Alter3 execute these actions.
Researchers have tested Alter3 with GPT-4 in various scenarios, including tossing a ball, eating popcorn, and playing air guitar. Alter3 can learn from its own memory and human responses. If the robot does something that elicits a positive reaction from humans, it will remember and attempt to replicate it in the future. This ability is comparable to how newborn babies imitate their parents’ expressions and gestures. Alter3 has also been programmed to exhibit humor and personality in its actions. In one instance, the robot pretends to eat popcorn, only to realize it belongs to someone else, leading to a surprised and embarrassed expression and an apologetic gesture.
The research team believes that Alter3 represents a breakthrough in robotics and artificial intelligence, showcasing how large language models can bridge the gap between natural language and robot control. This development opens up new possibilities for human-robot collaboration and communication, as well as the creation of more intelligent, adaptable, and personable robots.
The paper detailing the research, titled From Text to Motion: Grounding GPT-4 in a Humanoid Robot ‘Alter3,’ was authored by Takahide Yoshida, Atsushi Masumori, and Takashi Ikegami, and is available on the preprint server arXiv. The team hopes that their work will inspire further exploration and development in this field, ultimately leading to the creation of robot friends that can understand and relate to humans on a deeper level.
Alter3’s integration of natural language processing and robotics demonstrates how these technologies can combine to create remarkable interactions. By utilizing GPT-4, the robot can perform various tasks and behaviors based on verbal commands, eliminating the need for extensive programming or manual control. This capability also allows Alter3 to learn from its experiences and human feedback, and express a sense of humor and personality. The development of large language models has the potential to significantly enhance the fields of robotics and artificial intelligence, bringing us closer to the reality of having robots that can connect with us and provide entertainment.
Do you find Alter3 and its abilities intriguing? Would you like to have a robot like Alter3 as a companion? Share your thoughts and opinions with us.
