This is a conversion tool written in Java for converting OSM and Berlin Road Inspection GML (Berliner Straßenbefahrung) data into the Apollo OpenDRIVE format.
Apollo Auto (GitHub) is an open source autonomous driving platform using a derived version of the OpenDRIVE format developed by Baidu.
Compiled binaries can be found in the release section on the right.
The JOSM Apollo OpenDRIVE plugin can be found here
If you are using one of compiled release binaries you can skip this section and continue with getting the data for the conversion.
The source code can either be loaded into an IDE or be compiled via command line. The project has the following dependencies:
- cts-1.5.2
- google-collections-1.0-rc2
- jaxb-core-2.3.0
- jaxb-impl-2.3.0
- jaxb-xjc-2.3.0
- jts-core-1.17.1
- lucene-core-8.7.0
- lucene-spatial-8.4.1
- slf4j-api-1.7.36
- stax-utils-20070216
The conversion process requires at least an OSM file to create a basic road network. Additionally one or more Berlin Road Inspection GML files can be provided to enhance the quality of the resulting Apollo OpenDRIVE HD map. This section gives a quick overview on how to obtain the necessary data from the two data sources.
For OSM data, various sources, either official or through third-party services, exist, to download map data in different file formats. The conversion tool currently only supports the .osm file format, which is the most commonly used file format when dealing with OSM data. Depending on the scale of the desired area, different services exist to provide this file format.
To get OSM data of a geographically limited area the export function of the OSM website can be used. This function can be found in the menu located in the upper left corner of the OpenStreetMap website.
Here an area can be selected with the coordinates of the current view being the default value. The export can then be started via the "Export" button and the resulting file will start downloading after a short while. This method will only allow export into the .osm format.
The OSM format is derived by the XML standard. Here is an example of an OSM file downloaded from the official OSM website:
Data for the Berlin Road Inspection is publicly available on the official FIS-Broker of the government of Berlin. It is found under the category "Verkehr" and the application "Straßenbefahrung 2014".
A detailed description of all available data can be found here (PDF).
From there, various services providing data on the different feature layers can be accessed. This includes a WMS for image data and WFS services for each feature of the data set. Currently, only road markings ("Fahrbahnmarkierung Linie") and curbstone ("Bordstein") features are supported by the conversion tool.
An overview of the available WFS can also be found here.
To download the respective feature in a given area, it has to be directly requested from the API via an HTTP request. This can be done directly in the browser.
This is an example of a request for the "Bordstein" feature in a bounding box with the corners at the coordinates 375183.00,5820910.04 and 377867.68,5818479.47:
https://fbinter.stadt-berlin.de/fb/wfs/data/senstadt/s_Bordstein?&VERSION=2.0.0&SERVICE=WFS&REQUEST=GetFeature&BBox=375183.00,5820910.04,377867.68,5818479.47&TYPENAMEs=s_Bordstein
The resulting file can the be saved via the browser. Below is a snippet of the response to the request above:
The conversion tool can be started via a command line. It takes the following parameters:
- File path to the .osm file
- File path to the output file
- (optional) File paths to the .gml files
The tool needs to be run with Java version 8. This is an example of a command to convert one OSM and one GML file with the Java executable manually selected:
'C:\Program Files\Java\jdk1.8.0_211\bin\java.exe' -jar .\apolloconverter-0.1.0.jar INPUT.OSM OUTPUT.XODR INPUT1.GML
A set of demo files, containing one OSM file and two GML files are provided with the source code in the "demo" folder.
All three files contain data of the same area in Berlin.
Example of the demo GML data:
! Currently the parameters can not be changed via launch arguments. This is planned in a future update. !
For the process of combining adjoined elements in the GML data, a buffer zone is used on the endpoints of geometries to detect if other geometries are in close range. The size of this buffer zone can be changed to include elements further away if you have problems with geometries not connecting or different units of the coordinate system. It can also be lowered if the settings include too many unintended elements. The default value is calibrated to the EPDG 4326 and the lane size of the German road network. It is currently set at 0.00001 degrees, or about 70cm.
The image below shows the default radius (red) between to lanes (gray).
Similar to the buffer zone above, a different radius is used when determining which road a GML geometry should belong to. A higher radius will increase the pool of possible roads in the calculation, but will also increase the calculation time. It should therefore be calibrated as low as possible. For the demo data set a radius of 0.003 degrees, or about 4m has shown good results.
The converter is currently limited to the first of three levels of HD maps, as it is only capable of constructing the basic connected road network and boundaries. This layer includes the logical road connections for planning algorithms in autonomous driving.
Work is currently in progress to lift the conversion process into the second level with inclusion of complete lane data. Current problems include wrong lane classification of geometries, especially when roads split.
The converter uses the Apollo OpenDRIVE format, which is a reduced version of the official ASAM OpenDRIVE format. It is lacking some of the general traffic elements such as those connected to railroad infrastructure and is generally reduced in detail on road information, lanes and geometry types:
The end product of the converter is an complete drivable level 3 HD map with the official Apollo Auto demo file serving as an example. The demo "Sunnyvale" file can be found in the compiled Apollo Auto suite; below is a screenshot of a detailed intersection contained in the data.
This HD map data is used by the Apollo Auto platform as one of many data sources in the different modules of the architecture. In combination with pre-recorded sensor data a "test-drive" of a specific region can be created to test and demonstrate the platform in that area. Examples of these data-sets can be found on the Apollo developer website and are described below.
The Daoxiang Lake demo data set contains a test drive with data from a range of sensors.
- Point clouds from a Velodyne HDL-64E LiDAR
- Image data from cameras
- IMU data from gyro and accelerometer
- Pose data from the vehicle describing its current position and orientation
- Parameter data of the sensors
This data set is not accompanied by an HD map.
The South Bay data contains several locations in southern California. Each location has data from sever of the following sources:
- LiDAR scans and point clouds
- Pose data from the vehicle
- GNSS/IMU estimated poses
- Image data from cameras
At least one data set (SunnvaleBigloop) contains HD map data. This data is also compatible with the Apollo OpenDRIVE JOSM plugin.