Swarm Intelligence: A Novel and Unconventional Approach to Dance Choreography Creation

Michele Braccini, Allegra De Filippo, Michele Lombardi, Michela MilanoCREAI@ECAI 2024

Introduction

Creating artificial creativity in the performing arts presents a unique challenge. While AI has successfully generated music and text, digital or robotic dance often feels "canned," lacking the spontaneity and fluid adaptation of human improvisation. Traditional symbolic AI approaches typically rely on centralized control and rigid planning, which often results in performances that lack novelty or real-time adaptability.

This paper proposes a decentralized alternative: leveraging Swarm Intelligence (SI) to generate emergent, unpredictable, and aesthetically surprising group movements. By shifting from top-down planning to bottom-up self-organization, the system produces complex global structures from simple local interactions, mimicking the natural coordination seen in social insects.

The Generative Pipeline

The proposed framework bridges the gap between raw audio data and collective movement through a multi-stage pipeline designed to translate musical features into spatial dynamics.

The multi-stage pipeline for generating swarm choreography from music features.
  1. Music Feature Extraction: Using the Mido library, the system processes MIDI files to extract fundamental features such as tempo (BPM), volume, and note density. It also calculates higher-level "composite" features, such as the Shannon entropy of notes within a musical bar, to capture the rhythmic and melodic complexity of the track.
  2. Feature Mapping: These musical features are mapped directly to simulation parameters. For instance, high-intensity musical segments might increase the speed or mutual attraction parameters of the agents.
  3. Swarm Choreography Engine: Built in NetLogo, the simulation integrates two primary SI algorithms:
  4. Particle Swarm Optimization (PSO): Guides agents toward "choreographic anchors"—spatial patterns like lines or circles that act as temporary goals.
  5. Ant Colony Optimization (ACO): Implements a pheromone-like mechanism where agents leave traces, encouraging others to aggregate and form clusters.
  6. Dynamic Fitness Landscape: The "dance floor" is modeled as a fitness landscape where agents "consume" the floor's value as they move. This prevents the dance from becoming static, forcing the swarm into a state of continuous reorganization.

Emergent Behavior and Results

The system demonstrates that complex choreography can emerge as a byproduct of simple interaction rules rather than explicit programming. One of the key findings is the swarm's ability to transition between different states of organization based on musical cues, moving from chaotic dispersion to structured coordination.

Comparison of agent states: (A) scattered distribution and (B) spontaneous aggregation into a cluster, illustrating emergent group behavior.

Qualitative results from prototype case studies highlight several strengths:

  • Spontaneous Aggregation: The swarm successfully demonstrates macro-behaviors like clustering and the formation of recognizable spatial figures without a central leader.
  • Robustness and Flux: The integration of "memory" and "noise" allows the swarm to maintain structure while remaining in a state of "metastable" flux, mimicking the continuous flow of real dance.
  • Musical Synchrony: The mapping between Shannon entropy and SI parameters creates a visual impression of the swarm "following" the emotional and dynamical arc of the music.

Conclusion and Future Directions

While the current framework provides a robust proof-of-concept, several challenges remain. The field currently lacks a "gold standard" metric for evaluating artificial dance, making objective assessment difficult. Future work will focus on developing automated evaluation criteria and refining the selection of choreographic patterns to better reflect musical intent. By expanding the variety of SI mechanisms used, the authors aim to create an even broader "bouquet" of possible choreographic behaviors, further blurring the line between biological and artificial creativity.

Summary updated

BibTeX
@inproceedings{braccini2024swarm,
  author = {Michele Braccini and Allegra De Filippo and Michele Lombardi and Michela Milano},
  title = {{Swarm Intelligence: A Novel and Unconventional Approach to Dance Choreography Creation}},
  booktitle = {CREAI@ECAI},
  year = {2024},
}
Figure