UQnet

Abstract

Predicting the trajectories of road agents is fundamental for self-driving cars. Trajectory prediction contains many sources of uncertainty in data and modeling. A thorough understanding of this uncertainty is crucial in a safety-critical task like auto-piloting a vehicle. We need to distinguish between the uncertainty caused by partial observability of all factors that may affect a drivers’ nearfuture decisions, the so-called aleatoric uncertainty, and the uncertainty of deploying a model in new scenarios that are possibly not present in the training set, the so-called epistemic uncertainty. In this paper we propose a new framework to systematically quantify both sources of uncertainty. Specifically, to approximate the spatial distribution of an agent’s future position, we propose a 2D histogram-based deep learning model combined with deep ensemble techniques for measuring both aleatoric and epistemic uncertainty by entropy-based quantities. The proposed Uncertainty Quantification Network (UQnet) employs a causal part to enhance its generalizability. Experiments on the INTERACTION dataset show that UQnet significantly improves the generalizability of the missing rate compared to the previous state-of-the-art. Further analysis shows that high aleatoric uncertainty cases are mainly caused by heterogeneous driving behaviors and unknown intended directions. Based on this aleatoric uncertainty component, we estimate the lower bounds of mean-square-error and final-displacementerror as indicators for the predictability of trajectories. Furthermore, we use the epistemic uncertainty to identify rare cases in the test set. Our results illustrate that domain knowledge on speed-dependent driving behavior is essential for adapting a model from low-speed to high-speed situations. Our paper contributes to motion forecasting with a new framework, that recasts this problem in terms of model generalisation, and puts forward methods to quantify the resulting uncertainty.