How safe are self-driving cars?

Safety has been the main ‘banner’ value for developing automated vehicles, but it is difficult to responsibly claim it. Still the current evidence is insufficient, and it remains to be proved that these vehicles will be safe in all situations. 

Automated vehicles can eliminate a number of human errors that contribute to traffic crashes –alcohol, distraction, fatigue, exceeding speed limits. On the other hand, human drivers will be expected in many cases to take over control, for example when a sensor fails, or the automated vehicle simply exists it design domain. Can we expect that human drivers can stay alert if an AI algorithm has been driving the vehicle for a long time? Can we expect that they can maintain their driving skills in the long term? – we know from aviation that these are challenges with automation, that can create new types of accidents.

Another safety challenge is whether self-driving vehicles can safely navigate through a complex environment e.g. in an urban area with conventional cars, automated cars, pedestrians, cyclists, but also scooters, wheelchairs etc.
Research shows there are two main factors that will determine automated vehicles safety. One is penetration rate: the more automated vehicles are deployed, the more experience drivers and pedestrians will have with these vehicles to be able to safely interact with them – so there will be a learning curve. The other one is trust: drivers and all road users should be aware of the true capabilities of these AI systems, and neither mistrust them nor over-trust them.