Tsukada Lab. Computer Network Group
← 発表論文一覧 ワークショップ等

3D maps distribution of self-driving vehicles using roadside edges

Masaya Mizutani, Manabu Tsukada, Yuki Iida, Hiroshi Esaki

Eighth International Symposium on Computing and Networking Workshops (CANDARW 2020), Okinawa, Japan, pp. 40–45, 2020年11月25日

3行要約

  • 自動運転車が走行中に必要とする点群地図を、路側エッジのキャッシュから配信する。
  • Autowareに実装し、地図を取得しながら点群を用いた自己位置推定を続けられることを確認した。
  • 帯域を変えた実験では、100 Mbpsでのダウンロード時間は最大698 msだった。

論文の図

図 1: 路側エッジからの地図配信
図 1. 路側エッジからの地図配信出典: Masaya Mizutani ほか、IEEE CANDARW 2020(著者原稿の原図) の図 1地図データが保管元から路側エッジのキャッシュを経由し、自動運転車へ届く流れが分かる。
図 2: 点群地図を読み込む車載ソフトウェアの構成
図 2. 点群地図を読み込む車載ソフトウェアの構成出典: Masaya Mizutani ほか、IEEE CANDARW 2020(著者原稿の原図) の図 2ダウンロードした地図が、Autowareの自己位置推定機能に渡る経路が分かる。

アブストラクト(原文)

Three-dimensional (3D) maps have become a shared digital infrastructure for autonomous vehicles, especially in urban areas. Point Cloud Data (PCD) maps are used for scan matching to enable self-localization. Autonomous vehicles need to maintain PCD maps along with the destination that is often decided on demand and to keep the PCD map updated. In this paper, we propose a system that delivers PCD maps cached at roadside edges in real time. We implement the system in Autoware, open-source software for autonomous driving. Subsequently, we evaluate whether the autonomous vehicle can simultaneously download the PCD map from its edge and enable self-localization. Our results show that autonomous vehicles can perform self-localization while downloading the PCD map from the edge server. Additionally, we measure the download time with variable bandwidth and examine the bandwidth in which the self-localization normally operates. In our results, the download time of the PCD map at 100 Mbps was 698 ms at maximum, and the possibility of downloading PCD maps through 4G communication is shown.

動画

引用方法

M. Mizutani, M. Tsukada, Y. Iida, and H. Esaki, “3D maps distribution of self-driving vehicles using roadside edges,” in Eighth International Symposium on Computing and Networking Workshops (CANDARW 2020), Okinawa, Japan, 2020, doi: 10.1109/CANDARW51189.2020.00021.

BibTeX

@workshop{Mizutani2020,
  title         = {3D maps distribution of self-driving vehicles using roadside edges},
  author        = {Masaya Mizutani and Manabu Tsukada and Yuki Iida and Hiroshi Esaki},
  url           = {https://hal.archives-ouvertes.fr/hal-02997520/document?.pdf https://youtu.be/JBuX3JKDVjA},
  doi           = {10.1109/CANDARW51189.2020.00021},
  year          = {2020},
  date          = {2020-11-25},
  urldate       = {2020-11-25},
  booktitle     = {Eighth International Symposium on Computing and Networking Workshops (CANDARW 2020)},
  pages         = {40--45},
  publisher     = {IEEE},
  address       = {Okinawa, Japan},
  abstract      = {Three-dimensional (3D) maps have become a shared digital infrastructure for autonomous vehicles, especially in urban areas. Point Cloud Data (PCD) maps are used for scan matching to enable self-localization. Autonomous vehicles need to maintain PCD maps along with the destination that is often decided on demand and to keep the PCD map updated. In this paper, we propose a system that delivers PCD maps cached at roadside edges in real time. We implement the system in Autoware, open-source software for autonomous driving. Subsequently, we evaluate whether the autonomous vehicle can simultaneously download the PCD map from its edge and enable self-localization. Our results show that autonomous vehicles can perform self-localization while downloading the PCD map from the edge server. Additionally, we measure the download time with variable bandwidth and examine the bandwidth in which the self-localization normally operates. In our results, the download time of the PCD map at 100 Mbps was 698 ms at maximum, and the possibility of downloading PCD maps through 4G communication is shown.},
  keywords      = {ITS},
  pubstate      = {published},
  tppubtype     = {workshop}
}