CooperationCaptcha

Abstract

In the emerging field of automated vehicles (AVs), the many recent advancements coincide with different areas of system limitations. The recognition of objects like traffic signs or traffic lights is still challenging, especially under bad weather conditions or when traffic signs are partially occluded. A common approach to deal with system boundaries of AVs is to shift to manual driving, accepting human factor issues like post-automation effects. We present CooperationCaptcha, a system that asks drivers to label unrecognized objects on the fly, and consequently maintain automated driving mode. We implemented two different interaction variants to work with object recognition algorithms of varying sophistication. Our findings suggest that this concept of driver-vehicle cooperation is feasible, provides good usability, and causes little cognitive load. We present insights and considerations for future research and implementations.