Image-based assessment of road network readiness for automated driving

Abstract

Automated Driving (AD) is expected to deliver various benefits beyond those possible with manual driving for transport systems and the environment, yet there are many uncertainties with respect to the development path of AD to full automation. (SAE International, 2016) defines five levels of vehicle automation summarized in Figure 1. Automated driving system (ADS) can take over more driving tasks at higher automation levels until finally at level 5, ADS can handle the full range of driving complexity and it is feasible in all driving modes. However, the transition period to full automation might be long and full of uncertainties. Two incremental paths toward full automation have been observed so far. (CPBR, 2015) describes them as “something everywhere” and “everything somewhere”. Most traditional car manufacturers are embracing “something everywhere” path, i.e., gradually improving ADS in existing vehicles and shifting more driving tasks from the driver to ADS over time. Then the user is responsible for using the ADS wisely. This is also consistent with SAE automation levels. The other alternative, which was recently adapted by Google, involves aiming at full automation within a limited domain (e.g., only certain road types) and expanding this domain to more road types and more complex driving situations. This means the absolute ADS autonomy can only be realized in specific conditions. Then the challenge is to define those conditions specifically. For both paths, infrastructure is a defining factor. It can either facilitate or prevent higher automation capabilities. During the transition period to full automation, safe operation of levels 3-4 at their full automation capacity will highly depend on the type of infrastructure they encounter. For road authorities it is important to know how ready the road infrastructure is for safe automated driving. However, the academic literature and the field reports do not offer sufficient information to answer this question. (Farah et al., 2018) point out numerous knowledge gaps regarding infrastructure for AD. Therefore, we embarked on providing some insight into the matter via an expert workshop.