AutonoWeather: Enabling autonomous driving in winter conditions through optimized road weather interpretation and forecast

The goal of the project is to reduce road accidents by making autonomous vehicles more capable at operating in winter conditions, such as found in Norway. The current generation of self-driving cars, or driver assistant systems, do not contain the intelligence that is required to recognize slippery roads. Most cars that are sold in Norway today contain Advanced Driver Assistance Systems (ADAS) that aim to reduce road incidents. Examples are Autonomous Emergency Braking (AEB) and Lane Keeping Assist (LKA). However, a notorious limitation is that such systems do not function properly in winter conditions, and under certain circumstances can even increase the chance of serious incidents. As a consequence, it is believed that introducing self-driven cars on winter conditions will increase the risk of fatalities, when compared to only having (experienced) human drivers. It is believed that by providing these autonomous systems with the intelligence to determine the slipperiness of the road, the performance can greatly be improved, and the risk of fatalities can be reduced. The primary objective of the proposed study is to develop an accurate and affordable method for road-friction estimation in real-time. Such estimates are established using a novel combination of road weather models and car-mounted environmental sensors. Besides improved road safety, the proposed technology offers the potential for environmental benefits through intelligent route planning, which can reduce CO2 emissions, and offer optimized winter road maintenance, which reduces the need for chemicals that are harmful to the environment. A Road Weather Forecast solution was developed, where, using data from weather stations and/or from car sensors, combined with weather forecasts, it is possible to predict the condition of any road in Norway. In addition, a sensor box with environmental sensors that can be installed in cars was developed. With the data from these sensors, the road weather forecast is improved for the region where the car is driving and forecast models can be improved. A road condition detector using Machine Learning and Computer Vision was also developed. There, camera images and radar data are processed to estimate if the tracks have snow or asphalt. Finally, from the camera images, friction is also estimated using Machine Learning models.

The AutonoWeather project developed a road weather forecast solution which uses meteorological models, road weather stations measurements, weather forecast and meteorological sensors mounted on cars. This product can be accessed via an API or a web app, where the user can click on any road in Norway, choose the forecast period and obtain the road weather forecast. Another solution to detect the condition of the road in real time was developed using Machine Learning and Computer Vision, where it is possible to detect if the road in front of the car has asphalt or snow. In addition, a friction estimate solution based on camera images and radar measurements was developed. Finally, a Machine Learning solution was developed to improve the weather forecast with data from the sensors mounted on cars. The road weather forecast API is a prototype ready to be used by drivers to plan their journey or by the road authorities to plan maintenance measures. The other solutions, such as the friction estimation and the contaminant detection, are in proof-of-concept level and need further investigation.

Yearly an estimated 135 000 accidents occur on European roads, of which regrettably 25 500 people lose their lives. Most cars that are sold in Norway today contain Advanced Driver Assistance Systems (ADAS) that aim to reduce road incidents. Examples are Autonomous Emergency Braking (AEB) and Lane Keeping Assist (LKA). However, a notorious limitation is that such systems do not function properly in winter conditions, and under certain circumstances can even increase the chance of serious incidents. It is believed that by providing these autonomous systems with the intelligence to determine the slipperiness of the road, the performance can greatly be improved. This project proposes the development and evaluation of such a method. In order for this solution to be of practical use it needs to be (1) robust, (2) reliable, and (3) affordable. Considering the available (and competing) technologies, it is believed that weather forecast models offer a high potential for complying with these requirements. However, the current state of technology is not sufficient for these models to be suitable for direct friction estimation. More specifically, the accuracy of weather forecast models are limited due to the fact that the meteorological weather stations are stationary, which are few and have large distances in between. To overcome these limitations this project proposes a novel method that uses a suite of car-mounted environmental sensors combined with advanced weather modelling. In essence the car becomes a moving weather station, and thereby continuously obtains the most accurate and local weather data. Besides improved road safety, the proposed technology offers the potential for environmental benefits through intelligent route planning, which can reduce CO2 emissions, and offer optimized winter road maintenance, which reduces the need for chemicals that are harmful to the environment.