Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generation of Training Data from HD Maps in the Lanelet2 Framework

Fabian Immel, Richard Fehler, Frank Bieder, Christoph Stiller

TL;DR

The paper tackles the lack of a unified HD map framework that can both drive automated perception and generate training labels from maps. It introduces lanelet2_ml_converter, an extension to the Lanelet2 framework, implementing compound labels, traceability to map elements, and a real-time data-generation pipeline with 3D/6D pose support to serve both offline training and online inference. The authors provide a detailed implementation, define requirements for modern HD map frameworks, and validate the approach through application examples inspired by MapTR and OpenLaneV2, with open-source code integrated into Lanelet2. This work enables end-to-end map-based driving and learning workflows, offering a standardized, high-performance backbone for learning from HD maps and enabling online fusion and topology inference in real-world stacks.

Abstract

Using HD maps directly as training data for machine learning tasks has seen a massive surge in popularity and shown promising results, e.g. in the field of map perception. Despite that, a standardized HD map framework supporting all parts of map-based automated driving and training label generation from map data does not exist. Furthermore, feeding map perception models with map data as part of the input during real-time inference is not addressed by the research community. In order to fill this gap, we presentlanelet2_ml_converter, an integrated extension to the HD map framework Lanelet2, widely used in automated driving systems by academia and industry. With this addition Lanelet2 unifies map based automated driving, machine learning inference and training, all from a single source of map data and format. Requirements for a unified framework are analyzed and the implementation of these requirements is described. The usability of labels in state of the art machine learning is demonstrated with application examples from the field of map perception. The source code is available embedded in the Lanelet2 framework under https://github.com/fzi-forschungszentrum-informatik/Lanelet2/tree/feature_ml_converter

Generation of Training Data from HD Maps in the Lanelet2 Framework

TL;DR

The paper tackles the lack of a unified HD map framework that can both drive automated perception and generate training labels from maps. It introduces lanelet2_ml_converter, an extension to the Lanelet2 framework, implementing compound labels, traceability to map elements, and a real-time data-generation pipeline with 3D/6D pose support to serve both offline training and online inference. The authors provide a detailed implementation, define requirements for modern HD map frameworks, and validate the approach through application examples inspired by MapTR and OpenLaneV2, with open-source code integrated into Lanelet2. This work enables end-to-end map-based driving and learning workflows, offering a standardized, high-performance backbone for learning from HD maps and enabling online fusion and topology inference in real-world stacks.

Abstract

Using HD maps directly as training data for machine learning tasks has seen a massive surge in popularity and shown promising results, e.g. in the field of map perception. Despite that, a standardized HD map framework supporting all parts of map-based automated driving and training label generation from map data does not exist. Furthermore, feeding map perception models with map data as part of the input during real-time inference is not addressed by the research community. In order to fill this gap, we presentlanelet2_ml_converter, an integrated extension to the HD map framework Lanelet2, widely used in automated driving systems by academia and industry. With this addition Lanelet2 unifies map based automated driving, machine learning inference and training, all from a single source of map data and format. Requirements for a unified framework are analyzed and the implementation of these requirements is described. The usability of labels in state of the art machine learning is demonstrated with application examples from the field of map perception. The source code is available embedded in the Lanelet2 framework under https://github.com/fzi-forschungszentrum-informatik/Lanelet2/tree/feature_ml_converter
Paper Structure (5 sections, 5 figures)

This paper contains 5 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Requirements for Modern HD Map Frameworks
  3. Implementation
  4. Application examples
  5. Conclusions and Future Work

Figures (5)

  • Figure 1: Usage and applications of the software module. Together with the module, a Lanelet2 map can be used as a unified format for both map-based highly automated driving and as labels for a variety of map perception and map-based learning tasks.
  • Figure 2: Visualization of compound labels and the elimination of overlaps. The overlapping compound road borders 2 and 4 are eliminated in favor of the larger compound road borders 5 and 6.
  • Figure 3: Traceability of labels in local instance labels. Every label is associated with the underlying map element, visualized here with the letters A to D. This association is maintained over local instance labels. Compound labels have traceability of the composing labels available as well.
  • Figure 4: Labels for road borders, lane dividers and lane centerline paths with different fixed point numbers per feature similar to the labels used in MapTR maptr and MapTRV2 maptrv2.
  • Figure 5: Labels for lanelet centerlines and their connectivity to successors and neighbours visualized as green edges similar to the OpenLaneV2 dataset openlaneV2.