Researchers have introduced MotionGPT - a unique model for joint processing of language and human movement. Their methods, based on discrete vector quantization (more about what this is here), transform 3D movements into "motion tokens." By modeling movements and text in a unified way, they consider human movement as a specific language. This allows for the application of prompting-based learning methods to enhance results in movement-related tasks.
The practical applications of MotionGPT can be diverse. It can be used in robotics to create more natural and smooth movements for robots, as well as in the entertainment industry, for example, for character animation in games or movies. Moreover, it can be used to create virtual trainers or even rehabilitation assistants after injuries.