Neural-Swarm: Decentralized Close-Proximity Multirotor Control Using Learned Interactions

TL;DR

Neural-Swarm introduces a decentralized control method for multirotor swarms that leverages a neural network to accurately model aerodynamic interactions, enabling close-proximity flight with improved stability and reduced errors.

Contribution

The paper presents a novel neural network-based approach for modeling complex multirotor interactions, enabling stable close-proximity swarm control beyond traditional safety distances.

Findings

Significantly reduces height tracking errors compared to baseline controllers.

Demonstrates generalization of the learned model to larger swarm sizes.

Achieves stable close-proximity flight in experimental tests.

Abstract

In this paper, we present Neural-Swarm, a nonlinear decentralized stable controller for close-proximity flight of multirotor swarms. Close-proximity control is challenging due to the complex aerodynamic interaction effects between multirotors, such as downwash from higher vehicles to lower ones. Conventional methods often fail to properly capture these interaction effects, resulting in controllers that must maintain large safety distances between vehicles, and thus are not capable of close-proximity flight. Our approach combines a nominal dynamics model with a regularized permutation-invariant Deep Neural Network (DNN) that accurately learns the high-order multi-vehicle interactions. We design a stable nonlinear tracking controller using the learned model. Experimental results demonstrate that the proposed controller significantly outperforms a baseline nonlinear tracking controller with up to four times smaller worst-case height tracking errors. We also empirically demonstrate the ability of our learned model to generalize to larger swarm sizes.