SO(2)-Equivariant Downwash Models for Close Proximity Flight

Multirotors flying in close proximity induce aerodynamic wake effects on each other through propeller downwash. Conventional methods have thus far fallen short of providing adequate 3D force-based models that can be incorporated into robust control paradigms required when designing and deploying dense flight formations. Thus, learning a model for these aerodynamic downwash patterns presents an attractive solution. However, given the computational cost and inadequacy of downwash field simulators for real-world flight settings, data collection for training is confined to real-world experimentation, enforcing the need for sample efficient methods. In this paper, we leverage the latent geometry (e.g., symmetries) present in the downwash fields to accurately and efficiently learn models for the experienced exogenic forces. Using real world experiments, we demonstrate that our geometry-aware model provides improvements over comparable baselines, even when the model is 1/35th the size and has access to a third of the training data.