Learning movement through human-computer co-creative improvisation

Computers that are able to collaboratively improvise movement with humans could have an impact on a variety of application domains, ranging from improving procedural animation in game environments to fostering human-computer co-creativity. Enabling real-time movement improvisation requires equipping computers with strategies for learning and understanding movement. Most existing research focuses on gesture classification, which does not facilitate the learning of new gestures, thereby limiting the creative capacity of computers. In this paper, we explore how to develop a gesture clustering pipeline that facilitates reasoning about arbitrary novel movements in real-time. We describe the implementation of this pipeline within the context of LuminAI, a system in which humans can collaboratively improvise movements together with an AI agent. A preliminary evaluation indicates that our pipeline is capable of efficiently clustering similar gestures together, but further work is necessary to fully assess the pipeline's ability to meaningfully cluster complex movements.