2021年8月16日Blog

Not Just Another HD Map: AMP Approach to HD Map Development

記事を共有する

Note: Woven Planet became Woven by Toyota on April 1, 2023.

マッピング画像上のAMPロゴ

By Erika Kamptner, Senior Product Manager

Developing high-definition (HD) maps is the Automated Mapping Platform (AMP) team’s contribution to the Woven Planet vision, “Mobility to Love, Safety to Live.” For our team, this means we are broadening our scope to include:

  • More Road Types — Mobility beyond urban or high-traffic areas that is inclusive of rural landscapes, including the roads less traveled

  • Growing Regions — Mobility ranging from current megacities like Tokyo and Los Angeles to future growing metropolises

  • Mobility Infrastructure — Enabling highly accurate maps that inform the development of transit infrastructure and other smart mobility solutions for communities

The AMP team is building a platform to enable the development of global HD maps to support automated vehicle and smart mobility solutions. We first introduced our approach to HD maps in April 2021, and how map providers like Dynamic Mapping Platform Co., Ltd. can leverage AMP technology to keep their maps as up-to-date as possible.

What is an HD map and how is it used? What are the traditional approaches to HD mapping?

Vehicles equipped with Automated Driving (AD) and/or Advanced Driving Assistance Systems (ADAS) use sensors such as cameras, Global Navigation Satellite Systems (GNSS), Radio Detection and Ranging (RADAR), and Light Detection and Ranging (LiDAR) to achieve varying levels of autonomy on the road. Due to some limitations with these sensors, many original equipment manufacturers (OEMs also known as vehicle manufacturers) rely on high-definition maps, or “HD maps,” as an additional virtual sensor to support AD/ADAS applications. HD maps are high-accuracy, three-dimensional (3D) maps that can be used in the vehicle to provide additional context/prior information to the vehicle. The HD map supports vehicle localization (i.e. Where is the vehicle now? What is its relationship to other objects?) and includes detailed semantic information for road features including road, lane, traffic signs, and traffic light features.

The traditional approach to creating HD maps can be time consuming and costly to maintain. HD map providers have varying methods to creating and maintaining their HD map, but the approach typically relies on a fleet of vehicles specially equipped with expensive sensors to produce a highly accurate HD map. One challenge to this approach is not only the operational cost of managing a fleet of specialized vehicles for data collection, but also the breadth, and scope of that data collection, which are often limited to urban highways/expressways.

How is AMP’s approach to HD map development different?

Understanding the conventional challenges, our team approaches HD map development with a unique perspective to enable “Mobility to Love, Safety to Live” at a global scale:

1. Our Origins in Space

According to the United Nations, it is projected that over 95% of population growth will occur outside North America and Europe in the next decade — Africa and Asia seeing the greatest growth rates. AMP wants to ensure we can provide HD map coverage to support AD and smart mobility solutions where growing populations and economies need it most. To support this mission, our HD map pipeline begins with high-resolution orthorectified satellite and aerial imagery.

Vector objects that are detectable from aerial view, such as lane network, road boundary, road marker, and stop lines, are extracted from imagery through a series of machine learning pipelines, from object annotation and segmentation to feature detection.

The resulting vector features serve as a basemap or “initial map” to the AMP HD map. This “initial map” not only covers urban metropolitan areas, such as Tokyo, but secondary and tertiary roads in rural areas, which are sometimes excluded from traditional HD maps.

道路(赤)、木(緑)、実線(青)、車線マーカー(黄)の注釈付き衛星画像

Fig 1. Annotated satellite imagery of road (red), trees (green), solid lines (blue), and lane marker (yellow)

By using this satellite imagery extracted features as the base layer for our HD map, we can provide wider coverage than what would typically be achievable with just specialized fleet vehicles.

2. Where there’s a car, there’s a way a.k.a. The Power of Crowdsourcing

Next, this satellite-based base map is aggregated with features extracted from anonymized crowdsourced vehicle probe data, including from Toyota vehicles. Extracted features include high-accuracy semantic lane boundary information, traffic signage, and traffic light features. By using a high volume of crowdsourced data, the AMP HD map maintains a high rate of “freshness.” Another way to think about “freshness” is how closely a map feature represents the real-world at any given time. The crowdsourced approach to maintaining the map ensures that updates are made continuously and in near real time, as opposed to over set intervals, as is the case with some HD map provider approaches.

生成された道路境界(青)、車線マーカー(黄)、案内標識(ピンク)、ポール(オレンジ)、信号機(紫)を含む地図

Fig 2. Generated map includes road boundaries (blue), lane markers (yellow), guide signs (pink), poles (orange), and traffic lights (purple).

3. A(MP) Perfect Fit

Today, OEMs support several AD/ADAS applications from guiding your car to a parking spot to ensuring vehicles can safely traverse curved roads. In the smart mobility space, use cases span beyond the road to areas of pedestrian and micro-mobility. These varying applications and use cases require different features, attributes, and even varying required map accuracy levels. We believe that a single HD map can’t meet the varying needs of our customers’ use cases and so we want to put that customization into the hands of our users, allowing OEM AD/ADAS or smart mobility developers to only use what they need and deliver this data to them via standard services. The benefit to our users not only simplifies the complexity of the HD map for AD or smart mobility applications, but reduces map size and thus streaming costs for our customers.

What’s next for AMP?

Our vision is “Mobility to Love, Safety to Live” and this year we are realizing these initiatives through thrilling new AMP products and services. Our work supports AD/ADAS or smart city mobility developers to build custom HD map and geospatial solutions. From uploading and discovering data assets of interest, to processing and analyzing that data, and embedding data into customer solutions, we are always moving closer to our goals.

Join us! 👋

We are actively hiring exceptional talent for the AMP team. If you are interested in working with us in our brand-new office in Nihonbashi, Tokyo, please have a look at our current open positions!