Exploring Real World Map Change Generalization of Prior-Informed HD Map Prediction Models

Samuel M. Bateman,Ning Xu,H. Charles Zhao,Yael Ben Shalom,Vince Gong,Greg Long,Will Maddern

2024-06-05

Abstract:Building and maintaining High-Definition (HD) maps represents a large barrier to autonomous vehicle deployment. This, along with advances in modern online map detection models, has sparked renewed interest in the online mapping problem. However, effectively predicting online maps at a high enough quality to enable safe, driverless deployments remains a significant challenge. Recent work on these models proposes training robust online mapping systems using low quality map priors with synthetic perturbations in an attempt to simulate out-of-date HD map priors. In this paper, we investigate how models trained on these synthetically perturbed map priors generalize to performance on deployment-scale, real world map changes. We present a large-scale experimental study to determine which synthetic perturbations are most useful in generalizing to real world HD map changes, evaluated using multiple years of real-world autonomous driving data. We show there is still a substantial sim2real gap between synthetic prior perturbations and observed real-world changes, which limits the utility of current prior-informed HD map prediction models.

Robotics,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper aims to address the issue of high-definition (HD) map construction and updating in autonomous vehicles. Specifically, the focus of the research is on how to train models to predict high-quality online HD maps to support safe autonomous driving deployment. Although modern online map detection models have made some progress, their performance in practical applications is still unsatisfactory, especially in terms of prediction accuracy over long distances.

The paper explores the following two core questions:
1. Can online mapping models trained with synthetic perturbation prior information generalize to real-world map changes?
2. How does the consideration of various prior noise models and the mixed use of these noise models during training affect the generalization performance of these online mapping models to real-world map changes?

Through large-scale experimental studies, the researchers evaluated the performance of different types of synthetic perturbations on the models when faced with real-world map changes. They found that although the models demonstrated good denoising capabilities on synthetic data, there is still a significant gap when dealing with complex real-world changes. This indicates that current HD map prediction models based on prior information still have limitations in coping with map changes in real environments.