Design of music choreographies for multi-robot fleets via Machine Learning Language Models

Technology and arts are frequently juxtaposed as opposing fields, however this dichotomy fails to capture how their intersection has the potential to be a space where innovation meets imagination, unlocking a realm of exciting possibilities. The transformative power of technology can be harnessed to elevate artistic expression to new heights, and the creative impulses of artists can inspire technological innovation in turn. This thesis sets out to explore this convergence in the field of quadrotor dance, focusing on the development of a methodology to accurately model the dynamics of a small quadrotor with a linear system. The system is intended for use in an optimization-based navigation technique, enabling a swarm of quadrotors to fly safely to the rhythm of music. The process of system identification has been carried out using two different simulation framework: Gym-PyBullet-Drones and CrazyChoir. To achieve synchronization it is necessary to extract significant time instants from musical tracks, and this requires a pre-processing phase aimed at identifying the placement over time of the high-volume notes. From a visual standpoint, the best way to portray synchronization is to have the quadrotors perform a change in direction in correspondence of the aforementioned time instants. Given the growing attention to the role of artificial intelligence in art fields, it was thought to be of interest to test the artistic capabilities of a Large Language Model, by letting it design trajectories for each quadrotor in the swarm based on the information extracted in the pre-processing phase.