Both self-driving vehicles and driving assistance systems require a large amount of information to function properly. At first, we might think that the advanced sensors mounted on self-driving vehicles are already working well enough to know all the information about their environment. However, this is only partially true. In spite of the fact that vehicles can sense their environment accurately, they also need other sources of help to perform some of their functions. Self-driving vehicles also need maps - just like us drivers - for route plans, but traditional maps are often inaccurate and contain significantly less information than so-called High Definition Maps. HD maps are usually based on laser measurements, thus the accuracy of each road element can be around 2 centimeters. An HD map involves not only the route of the road and the number of lanes lying on it but also all their geometric features and their position in the coordinate system of the world. Furthermore, the HD map includes road signs, traffic signs, and various props lining the road. If all of this information is available to a vehicle, its localization can be immensely improved. A conventional GNSS system could determine an error up to 10 meters in the position of a vehicle, which is no longer acceptable for a self-driving car. However, this information is already sufficient to determine a smaller map section that describes the immediate environment of the vehicle. On this smaller map, the vehicle can already position itself with its sensors due to the information carried by the map. Besides, the vehicle can use the HD maps to know the speed limit data, the traffic light system, the location of pedestrian crossings and bike lanes, all of which can help with route and path planning.
Our goal is to properly support research in this direction as well, therefore we created an HD map of the selected modules of the ZalaZONE proving ground. The two track elements selected are the High-Speed Handling course and the Dynamic Platform. The latter consists of both multiple lane sections and intersections with lane painting. These sections can be used sufficiently during either localization or path planning. These maps are available for free under the downloads tab to provide useful assistance for as many of our partners and researchers as possible. These materials accessible for download include the high-resolution and high-precision Lidar point cloud which the maps are based on, recorded with the Lecia Pegasus 2 system. We have implemented our HD maps in the increasingly widespread OpenDRIVE format in the automotive industry. We also created a GeoJSON vector map from the OpenDRIVE format and generated a 3D fbx (filmbox) model that can be displayed for the more common 3D graphics engines. Due to their derivation from high-precision OpenDRIVE maps, they are also completely identical to reality, including the geometry of the road network, longitudinal and transverse elevation values, and road markings as they are drawn from the point cloud.
Many simulation software is used in the process of developing autonomous functions, so we are also working on producing formats that can be used immediately when integrated into different software. At present, a few simulation software is not able to convert the OpenDRIVE format to its own format with all its features, so we also created the IPG Carmaker and SUMO models of the two processed modules. These are also based on the OpenDRIVE model, but any shortcomings were already fixed with the help of the target software.
Self-driving vehicles require a wealth of driving experience, some of which developers want to create with the help of simulators. To achieve the best possible synthetic result, the use of 3D graphics engines has also come forward, on which several simulators build their visual display. That is why we created the Unity and Unreal Engine models of the two-track modules, which are available in film box format and as a Unity custom package.
If you are also interested in this topic and would like to work with our HD maps, you can request free access by clicking on the downloads tab, where you can also find out more about the available models and their content.
Written by Perception Team