Using HD maps to automatically control autonomous vehicles
Author: Vilim Šprajc
Mentor: izr. prof. dr. Matej Rojc univ. dipl. inž. el.
Degree: 1.
Date: september, 2020
DKUM: VILIM ŠPRAJC
HD maps
In the project we present in more detail how an autonomous vehicle can be aware of its actual position, how on the basis of this information the system can direct the vehicle in the environment, what technologies perform this, how successful they are, and what the main problems are. The universal acceptance and reduction of prices of Advanced Driving Automation Systems (ADAS) is bringing us closer to the age of self-driving vehicles, stage 5. Nevertheless, the ADAS system alone today is not sufficient for the realization of high-level autonomous vehicles. To solve one of the biggest challenges of autonomous driving – determining the exact position of the vehicle in real time – another key source is needed, namely high-resolution maps (HD maps).
Autonomous driving
An autonomous vehicle can recognize the environment in which it is located and automatically operate the vehicle with very little or no user control. Autonomous vehicles use various sensors to detect the environment, such as: radars, LiDAR, sonar, GPS, odometer, and IMU (‘Inertial measurement unit’). Sensory information is identified by advanced control systems to identify relevant traffic routes, obstacles and traffic signs. A car equipped with a fifth stage of autonomy is able to automatically control the vehicle in any situation, and automatically detects the surroundings in which it is located.
HD maps for autonomous vehicles
An autonomous vehicle needs a map in order to determine its current location and plan a route, HD maps actually allow for more advanced use due to accurate and richer information. For example, it helps to have information about the number of lanes on the road, and how the vehicle should be classified in order to reach the desired destination more optimally. Conventional maps are made for our needs, and are not particularly suitable for use in such systems. Unlike conventional maps, HD maps contain a greater variety of elements, such as speed limit data, details of buildings and other roadside structures, and so on. Also important is the much higher accuracy that HD maps provide even at the centimeter level.
HD maps formed from data from sensors
Generating HD maps using sensor data is one of the most widely used methods for generating maps. These vehicles usually include the following sensors. The GPS / IMU sensor records the spatial coordinates that are the basis for creating road routes in maps. To achieve good map accuracy, we need very accurate GPS. The vehicle is also equipped with one or more cameras placed in different directions. They are used to detect and locate objects around the roadway and also to detect road characteristics such as road markings (lane markings, other markings). LIDAR is used together with cameras to localize objects and other details of the vehicle’s surroundings.
Contents of the HD maps
HD map data is arranged in layers. Layering allows to independently design, create and import new data into HD maps. The layers are aligned and indexed to allow efficient reading of information for the current location of the autonomous vehicle and for the presentation of the surroundings. Basic data on the road network also given on freely accessible online maps are considered to be the lowest layer of such data. Each subsequent layer, however, adds additional information. The four most important layers are: basic map, geometric map, semantic map, priority map, real-time layer.
Shranjevanje HD zemljevidov
HD zemljevidi so nenehno uporabljani s strani sistema vozila, kot so sistem za zaznavanje, ter sistemi za načrtovanje in krmiljenje, tudi večkrat v sekundi. Modul za detekcijo sprememb se mora realno-časovno aktivirati, da preverja okolico vozila, npr. v čem se okolje razlikuje od že zabeleženih podatkov in preveriti, ali je razlika dovolj pomembna, da se jo uporabi v sloju realnega-časa (npr. gradbena dela, prometni zamaški, itd.). To informacijo se nato pošlje procesu robnega računalništva, in aktivira tudi proces aktivnega zbiranja podatkov, informacije se deli tudi z oblakom.
Diskusija in rezultati
