Visualizing Improvisation in LuminAI, an AI Partner for Co-Creative Dance
Duri Long, Lucas Liu, Swar Gujrania, Cassandra Naomi, Brian Magerko
Abstract
LuminAI is an art installation in which participants can improvise movements with an AI dance partner. In this practice work, we will present the LuminAI installation as well as two visualization tools that interactively demonstrate how the LuminAI agent reasons about movement using both bottom-up learned knowledge and top-down domain knowledge. Participants will first be invited to interact with the LuminAI installation, where they can improvise movement with an AI agent projected onto a screen. They can then see how LuminAI learns relationships between gestures by interacting with MoViz, a visualization in which participants can explore the agent's gesture memory and qualitatively compare the efficacy of unsupervised learning algorithms at clustering gestures. Finally, participants will be invited to interact with a third tool, where they can explore how LuminAI applies top-down domain knowledge to gesture reasoning. Participants will be able to interactively explore how LuminAI uses Laban Movement Analysis's conception of Space to analyze learned movements in terms of the geometric properties of Laban's icosahedron and manipulate these properties to transform and generate new movements. The two visualization tools both represent novel approaches to understanding and analyzing improvisational movement in creative domains.