Contact: Dümbgen, Frederike
Synopsis: Exploration and exploitation of drone noise
Level: MS semester project
Description: The more drones become omnipresent in everyday life, the more the unavoidable propeller noise is becoming an issue. Indeed, drone noise has been shown to be more unpleasant than car noise to many people [1], and thus recent research has focused on different ways to eliminate or reduce it.
In this project, we take a different angle to solve the problem and explore ways to make the drone sound more pleasant instead of reducing or eliminating the noise. As opposed to other common noise sources, drone noise is relatively coherent and has a strong harmonic structure. Thus, we can treat the drone as a musical instrument and explore the following questions:
- Can we characterize the set of maneuvers that sound good? For instance, a drone flying forward at a certain speed will generate sound with two different fundamental frequencies (one for the front propellers, one for the rear propellers). What speed leads to a pleasant sound interval?
- Can the drone fly melodies? What are the constraints on “flyable” melodies, i.e. melodies that a drone can fly without crashing?
- Can the drone auto-tune itself? We can use a simple forward model to predict the fundamental frequency given propeller commands. However, to “fly in tune” it might be necessary to implement a feedback loop.
Our idea to use the drone as a musical instrument is in contrast with previous works in which drones mechanically actuate musical instruments [2], external speakers emit sound based on drone movements [3] or drones move according to music [4].
Planning: In a first phase, the student will choose a simulation environment to fly the Crazyflie in simulation and test and understand different flight maneuvres. Potential solutions are (incomplete list)
- Crazyswarm simulation and control environment: https://crazyswarm.readthedocs.io
- CrazyS: https://github.com/gsilano/CrazyS
- gym-pybullet-drones: https://github.com/utiasDSL/gym-pybullet-drones
In a second phase, the student will augment the simulation environment to predict the sound of the drone(s). This will allow us to answer above questions in simulation. Finally, the created melodies and algorithms can be deployed on a real Crazyflie drone, if time permits.
References:
[1] Christian, A & Cabell, R. (2017): Initial Investigation into the Psychoacoustic Properties of Small Unmanned Aerial System Noise. Retrieved from https://ntrs.nasa.gov/citations/20170005870
[2] Flying Robot Rockstars: https://www.ajournalofmusicalthings.com/drone-music-music-played-drones/, accessed on February 4th, 2021.
[3] Schoellig, A. P., Siegel, H., Augugliaro, F., & D’Andrea, R. (2014). So You Think You Can Dance? Rhythmic Flight Performances with Quadrocopters. In A. LaViers & M. Egerstedt (Eds.), Controls and Art (pp. 73–105). https://doi.org/10.1007/978-3-319-03904-6
[4] Wang, X., Dalal, N., Laidlow, T., & Schoellig, A. P. (2015). A Flying Drum Machine. Retrieved from https://www.dynsyslab.org/wp-content/papercite-data/pdf/wang-tr15.pdf
Deliverables: A technical report, a simulation environment for testing drone sound, and a live demo (if time permits)
Prerequisites: Good programming skills, knowledge of drone dynamics and audio is a plus.
Type of Work: 20% theory/research, 50% coding, 30% experimental validation.