The Relationship Between Drowsiness Level and Takeover Performance in Automated Driving

Abstract

Although automated driving systems can perform dynamic driving tasks, at its lower levels, human drivers must still take over control of the vehicle whenever a Take Over Request (TOR) is issued. Human factors such as drowsiness may affect driver performance in responding to TOR. Results of research on the effects of different drowsiness levels on driver takeover performance appear inconsistent, for two possible reasons: 1. Some studies triggered TORs after a predefined duration of driving, while others triggered TORs based on a certain level of drowsiness. 2. Differences in drowsiness levels, which may have nonlinear effects on takeover performance, were not adequately considered. To investigate these inconsistencies, this experimental study recruited 40 participants and adopted a repeated-measures design. A total of 1436 available datasets, including a series of takeover performance measures and drowsiness measures, were collected. Analyses based on driving duration as well as drowsiness levels produced the following results: 1. Driving duration had a significant effect on drowsiness level but not on takeover performance. 2. Although mediumly drowsy drivers tended to sacrifice takeover quality for a fast reaction to ensure safety, highly drowsy drivers reacted significantly more slowly to TOR and were not able to maintain a safety margin comparable to that of drivers who were not highly drowsy. These findings have important implications for both researchers who are developing experimental studies to examine the effects of drowsiness on takeover performance and designers who use this information to design driver assistance systems.