Abstract
This paper aims to design quadrotor swarm performances, where the swarm acts as an integrated, coordinated unit embodying moving and deforming objects. We divide the task of creating a choreography into three basic steps: designing swarm motion primitives, transitioning between those movements, and synchronizing the motion of the drones. The result is a flexible framework for designing choreographies comprised of a wide variety of motions. The motion primitives can be intuitively designed using few parameters, providing a rich library for choreography design. Moreover, we combine and adapt existing goal assignment and trajectory generation algorithms to maximize the smoothness of the transitions between motion primitives. Finally, we propose a correction algorithm to compensate for motion delays and synchronize the motion of the drones to a desired periodic motion pattern. The proposed methodology was validated experimentally by generating and executing choreographies on a swarm of 25 quadrotors.
Introduction
While large-scale drone shows have become popular for entertainment, they typically rely on slow, sparse vehicle movements primarily designed for LED visual effects. Pushing these performances toward highly dynamic motions—where the swarm acts as a single, fast-moving, deforming entity—presents significant technical challenges. High-speed flight in close proximity introduces severe inter-agent collision constraints, aerodynamic disturbances like rotor downwash, and system delays. Even minor lags can disrupt the visual illusion of a cohesive swarm or lead to collisions. This research provides a unified pipeline for designing, transitioning, and executing dynamic swarm behaviors by bridging artistic choreography with robust robotic control.
Methods
Swarm Motion Primitives
The framework defines global swarm motions using a mathematical formulation based on sums of periodic functions (sines and cosines). By adjusting spatial and temporal parameters, choreographers can generate complex patterns such as 3D propagating waves or rigid body rotations. This approach allows for intuitive design of nature-inspired movements without manually plotting individual drone paths.
Transition Trajectory Planning
To move the swarm between different patterns, the system employs a two-step transition planner. First, goal assignment is performed using the Hungarian algorithm. Rather than minimizing simple Euclidean distance, the authors utilize a "minimum snap" cost function (minimizing the 4th derivative of position), which ensures fluid, energy-efficient transitions. Second, candidate trajectories are refined using a directed collision graph to resolve near-misses while preserving the swarm's geometric symmetry.
Motion Synchronization
To address real-world mechanical and communication lags, the quadrotors are treated as linear systems. By mapping the frequency response via Bode plots, the system identifies the phase lag and amplitude attenuation at specific frequencies. The reference trajectories are then pre-compensated—artificially scaling amplitude and shifting phase forward—to cancel out delays without the need for high-overhead real-time communication.
Key Results
The methodology was validated using a swarm of 25 Crazyflie 2.0 quadrotors in a compact 5x5x2 meter volume.
- Dynamic Performance: The swarm reached peak velocities of 2.65 m/s and pitch angles of 32 degrees while transitioning between five distinct wave motions and rigid body rotations.
- Precision and Safety: The synchronization algorithm significantly reduced tracking errors, allowing drones to maintain tight formations with horizontal collision boundaries of only 0.28m.
- Planner Robustness: In simulations, the transition planner successfully found collision-free paths for 85% of 1,800 randomly generated motion primitive pairs.
Limitations and Future Work
The current trajectory planner faces numerical instability during very long transitions, which could be mitigated by using concatenated polynomials rather than single high-order functions. Additionally, the system currently handles aerodynamic effects like rotor downwash by simply enlarging vertical collision ellipsoids. Future work aims to incorporate more explicit aerodynamic modeling to allow even tighter formations and push the physical limits of the vehicles.
Summary updated