Visualizing Detection Algorithms of Highly Automated Vehicles: Creating Transparency to Support Remote Assistant's Understanding and Predictability?

Kurzfassung

The scalable implementation of highly automated driving systems (ADS, SAE L4) on German roads depends on the availability of a remote assistant. Although ADS are highly sophisticated, they still need support to overcome technical limitations. Reaching these limitations will lead to minimal risk manoeuvres (MRM), causing the vehicle to safely stop in traffic. Remote assistants (RA) provide high level support for these situations. However, the effectiveness of a RA's intervention depends on the RA's understanding of the system state. The understanding of these RAs can be improved with transparent system design that provides information about the ADS. However, the most efficient design to communicate the information is yet to be determined. This study investigates the effect on a RA's understanding by providing information about an ADS's visual detection. Different types of visualizations were used to highlight detected objects in the ADS's video stream to the RA. In an experimental online study, the influence of the visualizations on the understanding, predictability, and complacency of RAs was investigated. Participants experienced different situations where they saw one of three types of different visualizations in the vehicle's video streams (boxing vs. saliency mapping vs. combined). Results indicated no influence on understanding and predictability. However, results on complacency provide insights into future research possibilities. This may shed light on adequate design solutions to improve trust and complacency towards the ADS.