"Human" Problems in Semi-Autonomous Vehicles: Understanding Drivers' Reactions to Off-Nominal Scenarios

Research related to autonomous vehicles and to their implications for human-machine interactions is on the rise. Advanced Driver Assistance Systems have become increasingly popular in vehicles currently deployed on the market, making researchers wonder about potential risks in case of technology failures for drivers that become accustomed to the use of such technology. To further our understanding of such concern, this work looks at the currently available data from autonomous vehicles field testing that has been carried out in California from 2014 to 2017. Our examination includes both qualitative and quantitative analyses, respectively, based on (i) the type of response in terms of control takeover in off-nominal scenarios that led to collisions involving autonomous vehicles; and (ii) the time to takeover after disengagements of the autonomous technology that acts as "brain" of the vehicle, with the request to the human driver to regain control of the vehicle. Our findings include expected values for the response time, discussion of factors that affect dispersion, presentation of how to determine trust and experience effects in the data, as well as a careful comparison with state-of-the-art literature on the topic.