To ensure safe and predictive autonomous driving it is crucial to capture and to map a vehicles surroundings as precisely and reliable as possible. While the major tasks are arguably the detection of other road users, infrastructure and obstacles, (road-) weather as an influencing factor is mostly understudied in this field. Nonetheless, the reliable detection and prediction of detailed local and particularly unexpected weather conditions such as reduced road friction or reduced visible distance especially during winter, is essential to safe autonomous driving functions as well as to generally reduce the overall risks of weather-related accidents. Currently, wide-meshed road conditions are deduced from weather models which source from a network of road-side weather stations, synoptic stations and partially mobile measurements. The coverage, however, does not satisfy the high temporal and spatial resolution as required by autonomous driving functions.
In the project “Flotten-Wetter-Karte” (Fleet Weather Map) the AUDI AG, cooperates with the “Deutscher Wetterdienst” (national weather service of Germany) to investigate on the technical requirements for collecting floating car data and integrating it into (road-) weather models. The effort is aimed to increase the overall spatial and temporal resolution of the weather observation network and the weather forecast model itself. The meteorological dataset evaluated in this project is composed of readings from mostly mass-produced vehicle-based sensors including air temperature, relative humidity, precipitation intensity and supported by prototypical sensors for visible distance and road temperature. The effective enlarging of the observation network presumably allows decreasing the forecast step width from currently about one hour to five minutes near ground level.
The paper provides an overview on vehicle sensors for weather-related information. We present setups and methods on evaluating and calibrating vehicle sensor readings to render these applicable for meteorological data assimilation and inclusion into weather prediction models. Preliminary experimental results are provided from real test vehicles.