Machine Sensible Infrastructure under Nordic Conditions

The automotive industry, together with software and hardware developers, is taking more and more steps in developing sensors and artificial intelligence (AI) to see and feel the road. Connected and automated vehicles (CAVs) can lead to better traffic safety and a more efficient transport system. However, from the government’s perspective, there is a great need for more scientific knowledge on AI and CAVs because information about how good the automated systems in cars are is often not shared by the automotive industry. In the Nordic countries, there is a particular need for more scientific studies that examine the limitations of these types of systems under Nordic conditions, as well as to investigate whether the authorities need to set different requirements for road design and winter maintenance. In this project, we use our own research platform for automated driving and collect data from a Kia equipped with an NVIDIA drive-by-wire kit to gain knowledge about how machine-readable roads in Norway are, and whether new ways for the car to identify where it can drive can be explored. MCSINC uses experiments and testing as the basis for all project activities, and with this, technology is tested while bringing together stakeholders and identifying common issues. Non-technological issues include regulation, policy, governance, standards, and business models. Solving these challenges requires a connection between research fields such as social sciences, road design and planning, and computer science. The project has been working on a test section on the E39 over Hemnekjølen, which is a mountain pass. Here, snow poles have been measured using RTK-GNSS to have an accurate reference of their positions. These poles can potentially be used in an HD map. Algorithms for recognizing such poles have been developed, and we have then tested whether the car can recognize the poles using LiDAR and cameras. By using the measured poles on the E39, we have examined whether the car can use the poles to determine its location, both as support for GNSS and in the absence of GNSS. The results so far indicate that the algorithms are good at recognizing the snow poles, and that navigation towards poles has a positive effect on the car’s ability to position itself. This is still under development. Additionally, the project has been working on the general generation of HD maps based on LiDAR and camera technology, especially concerning road markings.

The vehicle industry and software and hardware providers are rapidly developing sensor systems and artificial intelligence (AI) methods for sensing the road environment. Connected and Automated Vehicles (CAVs) are argued to have a large potential for accelerating traffic safety and efficiency. There is a strong need for more open scientific studies publishing AI software and results on CAV technology which give valuable insight to road authorities, information which is not available from the vehicle industry today. In addition, in Nordic countries more scientific studies on the limitations of these technologies caused by Nordic conditions is needed, and exploring how these limitations may set other requirements for road design and winter maintenance We utilize our own research platform for automated driving, i.e. a vehicle with automated driving capabilities, and data from instrumented vehicles to gain knowledge on how to establish a machine sensible road environment in the Nordic region and explore how standards for road design and maintenance should be adjusted in this regard. The proprietary nature of vehicle technologies makes a research platform for automated driving particularly important for gaining scientific knowledge on how these systems work on existing infrastructure and in real traffic. We use pilot activities as the backbone of the project, and aim at solving technological and non-technological challenges simultaneously. The non-technological challenges include development of governance, regulations, policies, standards and business models. Solving these challenges requires a coupling between the traditional fields of human factors, road planning and design and the field of computer science.