The paper "Uncertainty Quantification of Car-following Behaviors: Physics-Informed Generative Adversarial Networks [PDF] "
by Zhaobin Mo and Xuan Di, accepted by 2022 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining's workshop UrbanComp (http://urban-computing.com/urbcomp2022/accept.html), won the best paper award (see attached for the certificate), announced on Aug 15, 2022 during the KDD conference in Washington DC.
This paper aims to characterize uncertainty in human driving behavior using physics-informed generative adversarial networks (GAN). We developed a "DoubleGAN" framework that encodes stochastic physics equations into GAN and introduces a second GAN, aiming to capture stochasticity arising from endogenous heterogeneity like behavioral randomness exhibited by individual drivers. This method is applied to a real-world dataset, NGSIM, which demonstrates superior performance over other baselines.
