Zusammenfassung
Motivation: In spite of technological advances, autonomous cars and driver assistance systems can occasionally fail (Levin, 2018). Therefore, it remains necessary for humans to be continuously vigilant for instances of automation failure (Parasuraman, Sheridan, & Wickens, 2000). Auditory warnings could assist by alerting users when there is a high likelihood of danger. In this work, we investigate how auditory warning cues can hasten emergency braking responses.
Objective: This study investigates the use of auditory warnings for potential rear-end collisions, in the case of a failure in the autonomous cruise control (ACC) which measures and keeps the correct distance to front cars. In particular, we investigate the benefit of using looming sounds (i.e., sounds that grow in intensity) over static intensity sounds. Unlike static sounds, looming sounds convey two pieces of information: First, they give information that something is approaching. Second, they provide an estimation of the time of the collision.
Background: Gray (2011) compared different auditory warning cues for braking situations in a realistic driving simulation that required full manual handling. He showed that looming sounds resulted in faster braking responses to potential collisions than sounds with unchanging intensities. Yet, it remains unclear if this difference is due to attentional redirection from other aspects of manual driving (e.g., lane-keeping) to emergency braking or whether looming sounds are especially suited for signaling an emergency braking response. Neuroimaging studies have shown that the brain processes looming stimuli preferentially (e.g. Bach, Furl, Barnes, & Dolan, 2015). Furthermore, looming sounds support the speedy detection of visual looming objects (e.g. Alink et al., 2012). It has been claimed that this preferential crossmodal benefit of looming sounds occurs because they signal approaching threat. Consistent with this claim, it has been shown that looming sounds induce selective activation of the amygdala (Bach et al., 2008).
Method: For this study, we tested 20 volunteers (13 females, 7 males) aged between 20 and 32 (M=25). Our study was conducted in a driving simulator that was programmed to drive with ACC (Fig. 1). The participants’ only task was to watch the scene and to step on the braking pedal when a rear-end collision with a lead vehicle could occur. Possible crash situations were occasionally cued (66.7%) by either a static-intensity sound or a looming sound. Sounds could also be presented, in absence of danger (false alarms, 20 %).
We measured braking times as well as the event-related potential (ERP) evoked by the auditory cues. To this end, the electroencephalogram (EEG) was recorded via 63 active electrodes while participants performed the task. The electrodes were distributed according to the international 10-20-system with the reference electrode positioned at FCz and the ground electrode at AFz. The electro-oculogram activity was recorded by four electrodes around the eyes. EEG data was filtered, cleaned using the Artifact Subspace Reconstruction method and re-referenced to the average reference. Afterwards, we cleaned the EEG using an independent component analysis (ICA) to reduce artifacts without losing too much data. We hand-selected eight out of 30 clusters, which were likely to represent noise, for removal and back-projected the remaining EEG activity for analysis at the electrode level.
Results: Braking to potential crashes was significantly faster when either the looming (M=825 ms, SD=112) or the constant (M=845 ms, SD=123) sound were played compared to when no sound was presented (M=1100 ms, SD=112), F(2,36)=199.5, p<.001. Moreover, looming sounds generated faster braking than static sounds, t(18)=-2.46, p=.036 (Fig. 2). There was no significant difference in false alarms between the looming and the constant sound condition, t(18)=.57, p=.58. A mass-univariate analysis of the ERP data revealed differences in the ERP in the time-interval between 240 and 280 ms after sound onset (Fig. 3a, b): The static sound evoked a larger fronto-central N2 and a larger left parietal P2 amplitude compared to the looming sound (Fig. 3). We found no differences in early, perceptual components or in the readiness potential.
Discussion: The behavioral data indicates that looming sounds speed up braking reactions, even if no redirection of attention from driving to emergency braking is required. In other words, looming sounds result in fast emergency braking during manual vehicle handling (Gray, 2011), as well as partially automated driving (i.e., ACC).
The EEG data shows that this looming benefit emerges as early as 200 ms after sound onset, as indicated by the smaller N2 after the looming sound compared to the static sound. We attribute this amplitude difference to the audiovisual incongruity between the looming visual stimulus and the static sound (Lindström et al., 2012). Specifically, we suggest that a looming sound which is consistent with a looming visual stimulus leads to faster braking times because it is easier to process. This is reflected by a smaller fronto-central N2. Furthermore, we speculate that the amplitude difference observed for the left parietal P2 probably derives from the same underlying neural mechanisms because both effects occurred within the same time window.
Further work is necessary to specify the underlying neural sources that could have given rise to this.
In summary, our results show that looming sounds benefit performance also when driving with ACC. Specifically, we find that it generated faster braking reactions. We believe that the reaction is facilitated by the faster processing of congruent audiovisual stimuli.
Limitations and outlook: The neural sources that give rise to the observed looming benefit remain unknown. A source estimation analysis or neuroimaging techniques with high spatial resolution (e.g., MEG, fMRI) will provide further insight. Furthermore, although driver inattention is the targeted real-world problem of our inquiry, our participants were instructed to attend the driving scene. Future extensions of this work could introduce a secondary, non-driving task to examine the looming benefit when drivers are no longer required to drive and thus do not pay attention to the driving. Finally, looming objects can approach one’s vehicle from more directions than only the front. Thus, it could be worthwhile to investigate whether looming sounds from the side could result in faster evasion maneuvers, or whether the looming benefit is exclusive to emergency braking only.
