Physics-Informed Neural Networks for Trajectory Prediction and Uncertainty Quantification in Urban Air Mobility

This study presents a Physics-Informed Neural Networks (PINNs) model for trajectory prediction and uncertainty quantification in Urban Air Mobility (UAM) with a focus on electric vertical take-off and landing (eVTOL) vehicles. UAM faces challenges from variable urban conditions, demanding accurate, real-time trajectory predictions. By embedding physical laws into the network, the PINNs model ensures alignment with eVTOL dynamics, providing superior prediction accuracy and robust uncertainty quantification compared to traditional neural networks (NNs) and Gaussian Mixture Regression (GMR). Using NASA's UAM simulation data, we demonstrate that the PINNs model not only achieves high accuracy but also enhances interpretability and reliability for safety-critical UAM applications. This work highlights PINN's potential to improve UAM safety and efficiency, laying the groundwork for advanced aerospace predictive modeling.