View article

Autonomous driving in level 2 and level 3 autonomy has been adopted by multiple companies such as Tesla and BMW. Autonomous driving relieves the cognitive load of drivers while also challenges the relations between drivers and autonomous agents. Effective and safe driving calls for appropriate collaborations between drivers and autonomous agents. Especially in risky scenarios, how do autonomous agents and drivers behave? What is the proper way for drivers to perceive the scenarios? To answer these questions, this article turned to data from real scenarios and collected a broad range of autonomous driving videos during which micro accidents happened. Here the micro accidents refer to abnormal events that are risky but not fatal, such as sharp braking and red-light violation. The features of these scenarios were analyzed using machine learning methods. Drivers’ perceptions and reactions to these micro-accidents were additionally explored using crowdsourcing experiments. The results provided an initial exploration of critical variables that induce micro accidents, and revealed the features of drivers’ perception of micro accidents. Based on these results, this article discussed possible improvements in interactions between autonomous agents and drivers.