Nonlinear System Identification of Swarm of UAVs Using Deep Learning Methods

TL;DR

The paper tackles nonlinear system identification for planar UAV swarms, comparing non-deep-learning, standard deep learning, and Neural ODE approaches using time-series data from swarm simulations and evaluating with Mean Field Error (MFE). It finds that ML models (MLP, RNN, CNN) struggle to generalize across transient dynamics and varying initial conditions, performing best only when trained on steady-state data with identical initial conditions; Neural ODE with a physics-informed architecture improves robustness but remains challenged by bifurcations and hysteresis inherent to swarm dynamics. The study highlights Neural ODE as a promising direction for robust trajectory forecasting in multi-agent UAV systems, while calling for further research to fully capture nonlinear emergent behavior. Overall, the work guides method selection for swarm trajectory prediction and underscores the role of physics-informed, adjoint-based training in achieving generalizable dynamics models.

Abstract

This study designs and evaluates multiple nonlinear system identification techniques for modeling the UAV swarm system in planar space. learning methods such as RNNs, CNNs, and Neural ODE are explored and compared. The objective is to forecast future swarm trajectories by accurately approximating the nonlinear dynamics of the swarm model. The modeling process is performed using both transient and steady-state data from swarm simulations. Results show that the combination of Neural ODE with a well-trained model using transient data is robust for varying initial conditions and outperforms other learning methods in accurately predicting swarm stability.

Figures (14)

• Figure 1: MFE for Steady State Predictions
• Figure 2: MFE for Transient Predictions
• Figure 3: Prediction for Steady State
• Figure 4: Prediction for Transient
• Figure 5: RNN Unrolled Structure