Robust Monocular Localization in Sparse HD Maps Leveraging Multi-Task Uncertainty Estimation

Abstract

Robust localization in dense urban scenarios using a low-cost sensor setup and sparse HD maps is highly relevant for the current advances in autonomous driving, but remains a challenging topic in research. We present a novel monocular localization approach based on a sliding-window pose graph that leverages predicted uncertainties for increased precision and robustness against challenging scenarios and perframe failures. To this end, we propose an efficient multi-task uncertainty-aware perception module, which covers semantic segmentation, as well as bounding box detection, to enable the localization of vehicles in sparse maps, containing only lane borders and traffic lights. Further, we design differentiable cost maps that are directly generated from the estimated uncertainties. This opens up the possibility to minimize the reprojection loss of amorphous map elements in an associationfree and uncertainty-aware manner. Extensive evaluation on the Lyft 5 dataset shows that, despite the sparsity of the map, our approach enables robust and accurate 6D localization in challenging urban scenarios using only monocular camera images and vehicle odometry.

Uncertainty-aware Panoptic Segmentation

“…Here, the network is trained to collect parameters for a high-order distribution, the Dirichlet distribution in their case, from which the uncertainty of the prediction is computed. Petek et al [8] utilized this method in multitask learning setting to simultaneously predict semantic segmentation and bounding box regression uncertainties. In our approach, we build upon evidential deep learning to learn semantic segmentation, instance segmentation, and bounding box classification uncertainties.…”

“…In this context the research in sampling-free methods for uncertainty estimation is gaining interest. One such method is evidential deep learning, which is already being used successfully in classification [6], regression [7], and multitask learning settings [8].…”

TrackletMapper: Ground Surface Segmentation and Mapping from Traffic Participant Trajectories

“…To allow for robust and safe navigation, autonomously operating robots in urban environments are required to localize nearby traffic participants accurately [1,2,3] and classify ground surfaces robustly. While autonomous vehicles typically require a binary distinction between road and non-road surfaces, mobile robots operating in pedestrian spaces must crucially be able to distinguish between sidewalks, roads, and road crossings in order to navigate urban environments safely [4,5,6].…”