Takeover request (TOR) effects during different automated vehicle failures

Abstract

Research on driving automation has investigated the use of takeover requests (TORs) to warn drivers about automation failures that require their intervention. Such failures can occur with information in the environment that drivers can use to anticipate them (e.g., system-limit failures) or without such information (e.g., system-malfunction failures). There is a lack of research comparing the effectiveness of TORs prior to these different failure types. We conducted a simulator study with 19 participants to investigate whether the effect of a TOR on drivers’ monitoring and takeover performance differed by failure type. Drivers were trained on automation limits so that they could identify upcoming system-limit failures. We evaluated gaze behaviors starting from 6 seconds before the failure (corresponding to TOR onset in TOR drives and the equivalent point in no-TOR drives) until drivers took over. The effect of TORs on monitoring the roadway was significant only for system-malfunction failures, with participants looking more at the roadway in TOR drives compared to no-TOR drives. For system-limit failures, the TOR did not provide any benefit in terms of visual attention to the roadway, likely because participants were already looking at the roadway because they could anticipate the failure. However, having a TOR for system-limit failures was associated with faster takeover time than not having a TOR. Although the TOR may not have had monitoring benefits when environmental information was available, our findings suggest it was still useful as a confirmation of impending failures and a prompt to take over control of the vehicle.