Designing Movement Generation Models in Collaboration With Voguing And Dancehall Dancers

Paper Title

Designing Movement Generation Models in Collaboration With Voguing And Dancehall Dancers

Publication Info

• Topic area: AI-driven generative models for dance movement creation.
• Keywords: AI, generative models, dance, Voguing, Dancehall, motion capture, stylistic legibility, creativity support tools, participatory design.

Background and Problem

• Problem / challenge: Current AI dance movement generation models lack integration of dance practitioners' expertise and fail to capture the stylistic essence of specific dance genres, limiting their applicability in choreographic settings.
• Significance: Addressing this gap is crucial for creating tools that support dancers' creativity and preserve the unique stylistic conventions of non-institutionalized dance forms like Voguing and Dancehall.
• Motivation and related work: Previous research has focused on technical capabilities of generative models, often trained on generic datasets without practitioner involvement. Studies have highlighted challenges in aligning generated movements with dancers' expectations, especially for codified dance styles. This paper builds on these insights by involving practitioners in the design and evaluation process.

Solution

• Proposed approach: Development of an AI-based movement generation model tailored to Voguing and Dancehall styles, supported by the interactive visualization tool Korai.
• Novelty:
  1. Collaborative design of generative models with dance practitioners, emphasizing stylistic fidelity.
  2. Creation of Korai, a web-based tool for real-time visualization and interaction with generated movements.
  3. Empirical studies investigating dancers' responses to models with varying levels of stylistic legibility and physical realism.
• Procedure and key techniques:
  • Motion capture of curated Voguing and Dancehall sequences.
  • Iterative development of generative models using the Chor-rnn architecture.
  • Design and deployment of Korai for interactive exploration and visualization of generated movements.
  • Structured observation and walkthrough studies with dancers to evaluate models across fidelity levels.

Results

• Concrete findings:
  • Dancers preferred movements either highly faithful to their repertoire or highly novel and unfamiliar, rejecting medium-fidelity outputs as uninspiring.
  • High-fidelity models facilitated fluid improvisation and integration into Voguing and Dancehall styles.
  • Low-fidelity models sparked creativity despite glitches, while medium-fidelity models were perceived as abstract and difficult to interpret.
• Advantage over baselines: Direct collaboration with dancers ensured stylistic legibility and practical relevance, addressing gaps in prior generative systems trained on generic datasets.
• Experiments / evaluation:
  • Motion capture dataset: 35 minutes of curated Voguing and Dancehall movements.
  • Studies: Improvisation sessions, structured walkthroughs, and comparative structured observations.
  • Metrics: Qualitative thematic analysis of dancers' feedback and improvisation strategies.
• Limitations and future work:
  • Limited participation of dancers in model training and design.
  • Need for larger-scale studies across diverse dance styles and longitudinal settings.
  • Exploration of newer architectures to improve fidelity and realism.

Summary

This paper presents a long-term collaboration with Voguing and Dancehall dancers to design movement generation models tailored to their repertoire. The iterative process involved motion capture, model development using Chor-rnn, and the creation of Korai, an interactive visualization tool. Empirical studies revealed that dancers favored either highly faithful or highly novel outputs, highlighting the importance of stylistic legibility in codified dance styles. The findings suggest an "uncanny valley" effect in dance generation, where medium-fidelity models are perceived negatively. The research advocates for participatory approaches in AI design and highlights Korai as a tool to bridge gaps between AI development and embodied artistic practices.