Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LMT-Net: Lane Model Transformer Network for Automated HD Mapping from Sparse Vehicle Observations

Michael Mink, Thomas Monninger, Steffen Staab

TL;DR

This work proposes Lane Model Transformer Network (LMT-Net), an encoder-decoder neural network architecture that performs polyline encoding and predicts lane pairs and their connectivity and evaluates the performance of LMT-Net on an internal dataset that consists of multiple vehicle observations as well as human annotations as Ground Truth (GT).

Abstract

In autonomous driving, High Definition (HD) maps provide a complete lane model that is not limited by sensor range and occlusions. However, the generation and upkeep of HD maps involves periodic data collection and human annotations, limiting scalability. To address this, we investigate automating the lane model generation and the use of sparse vehicle observations instead of dense sensor measurements. For our approach, a pre-processing step generates polylines by aligning and aggregating observed lane boundaries. Aligned driven traces are used as starting points for predicting lane pairs defined by the left and right boundary points. We propose Lane Model Transformer Network (LMT-Net), an encoder-decoder neural network architecture that performs polyline encoding and predicts lane pairs and their connectivity. A lane graph is formed by using predicted lane pairs as nodes and predicted lane connectivity as edges. We evaluate the performance of LMT-Net on an internal dataset that consists of multiple vehicle observations as well as human annotations as Ground Truth (GT). The evaluation shows promising results and demonstrates superior performance compared to the implemented baseline on both highway and non-highway Operational Design Domain (ODD).

LMT-Net: Lane Model Transformer Network for Automated HD Mapping from Sparse Vehicle Observations

TL;DR

This work proposes Lane Model Transformer Network (LMT-Net), an encoder-decoder neural network architecture that performs polyline encoding and predicts lane pairs and their connectivity and evaluates the performance of LMT-Net on an internal dataset that consists of multiple vehicle observations as well as human annotations as Ground Truth (GT).

Abstract

In autonomous driving, High Definition (HD) maps provide a complete lane model that is not limited by sensor range and occlusions. However, the generation and upkeep of HD maps involves periodic data collection and human annotations, limiting scalability. To address this, we investigate automating the lane model generation and the use of sparse vehicle observations instead of dense sensor measurements. For our approach, a pre-processing step generates polylines by aligning and aggregating observed lane boundaries. Aligned driven traces are used as starting points for predicting lane pairs defined by the left and right boundary points. We propose Lane Model Transformer Network (LMT-Net), an encoder-decoder neural network architecture that performs polyline encoding and predicts lane pairs and their connectivity. A lane graph is formed by using predicted lane pairs as nodes and predicted lane connectivity as edges. We evaluate the performance of LMT-Net on an internal dataset that consists of multiple vehicle observations as well as human annotations as Ground Truth (GT). The evaluation shows promising results and demonstrates superior performance compared to the implemented baseline on both highway and non-highway Operational Design Domain (ODD).
Paper Structure (27 sections, 12 equations, 5 figures, 2 tables)

This paper contains 27 sections, 12 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Grid-based Map Construction
  4. Graph-based Map Construction
  5. Lane Mapping from Fleet Data
  6. Approach
  7. Problem Statement
  8. LMT-Net Architecture
  9. Loss Function
  10. Implementation
  11. Model Implementation
  12. Training Details
  13. Experiments
  14. Dataset
  15. Evaluation Metrics
  16. ...and 12 more sections

Figures (5)

  • Figure 1: The network architecture of LMT-Net consists of three main blocks. Example input polylines are shown on the left. A Polyline Encoder constructs polyline-level feature vectors from vectorized polylines and a Center Point Encoder generates Queries from center points. An encoder-decoder transformer module is used to construct a latent representation of the driven traces and observed lane boundaries. Two MLP-based prediction heads output lane pairs and connectivity, which form the lane graph. Example output for 4 center points is shown on the right, and the output for one is highlighted.
  • Figure 2: Two examples: (a) a highway scenario and (b) a non-highway scenario. Top shows raw traces in yellow. Middle shows input data with observed lane boundaries $O$ in green and driven traces $T$ in red. The bottom shows ground truth lane boundaries $L$ in blue. Observed lane boundaries can be incomplete (left) or noisy (right).
  • Figure 3: Illustration of baseline implementations for lane pair prediction.
  • Figure 4: Predictions of LMT-Net on various German road scenarios. Orange polygons depict areas formed by predicted lane pairs. Blue arrows show predicted lane connectivity.
  • Figure 5: Variants of polyline encoding: left is shared, right is type-specific.