Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mind the map! Accounting for existing map information when estimating online HDMaps from sensor

Rémy Sun, Li Yang, Diane Lingrand, Frédéric Precioso

TL;DR

This paper introduces MapEX, a novel online HDMap estimation framework that accounts for existing maps by encoding map elements into query tokens and by refining the matching algorithm used to train classic query based map estimation models.

Abstract

While HDMaps are a crucial component of autonomous driving, they are expensive to acquire and maintain. Estimating these maps from sensors therefore promises to significantly lighten costs. These estimations however overlook existing HDMaps, with current methods at most geolocalizing low quality maps or considering a general database of known maps. In this paper, we propose to account for existing maps of the precise situation studied when estimating HDMaps. We identify 3 reasonable types of useful existing maps (minimalist, noisy, and outdated). We also introduce MapEX, a novel online HDMap estimation framework that accounts for existing maps. MapEX achieves this by encoding map elements into query tokens and by refining the matching algorithm used to train classic query based map estimation models. We demonstrate that MapEX brings significant improvements on the nuScenes dataset. For instance, MapEX - given noisy maps - improves by 38% over the MapTRv2 detector it is based on and by 8% over the current SOTA.

Mind the map! Accounting for existing map information when estimating online HDMaps from sensor

TL;DR

This paper introduces MapEX, a novel online HDMap estimation framework that accounts for existing maps by encoding map elements into query tokens and by refining the matching algorithm used to train classic query based map estimation models.

Abstract

While HDMaps are a crucial component of autonomous driving, they are expensive to acquire and maintain. Estimating these maps from sensors therefore promises to significantly lighten costs. These estimations however overlook existing HDMaps, with current methods at most geolocalizing low quality maps or considering a general database of known maps. In this paper, we propose to account for existing maps of the precise situation studied when estimating HDMaps. We identify 3 reasonable types of useful existing maps (minimalist, noisy, and outdated). We also introduce MapEX, a novel online HDMap estimation framework that accounts for existing maps. MapEX achieves this by encoding map elements into query tokens and by refining the matching algorithm used to train classic query based map estimation models. We demonstrate that MapEX brings significant improvements on the nuScenes dataset. For instance, MapEX - given noisy maps - improves by 38% over the MapTRv2 detector it is based on and by 8% over the current SOTA.
Paper Structure (16 sections, 2 equations, 3 figures, 1 table)

This paper contains 16 sections, 2 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. What Kind of Existing Map Could We Use?
  4. HDMap Representation For Online HDMap Estimation
  5. MapModEX: Simulating Imperfect Maps
  6. Scenario 1: Only Road Boundaries Are Available
  7. Scenarios 2a and 2b: Maps Are Noisy
  8. Scenarios 3a and 3b: Maps Have Substantially Changed
  9. MapEX: Accounting for EXisting Maps
  10. Overview
  11. Base framework
  12. Our MapEX framework
  13. Translating Maps into EX Queries
  14. Map Element Pre-attribution for Training
  15. Experimental Results
  16. ...and 1 more sections

Figures (3)

  • Figure 1: We propose to use existing map information - even if inexact - to estimate better online HDMaps from sensor inputs. In doing so, we simplify the problem from generating maps using only sensors to re-using available existing maps aided by sensors.
  • Figure 2: Examples of HDMaps generated by MapModEX.
  • Figure 3: Overview of our MapEX method (see \ref{['sec:mapex']}). We add two modules (EX query encoding, Attribution) to the classic query based pipeline. Map elements are encoded into EX queries that are added to classic learned queries. These queries are then decoded using sensor data to yield map elements. Pre-attribution of known prediction to true map elements helps train stronger models.