Paper - Interrupting Drivers for Interactions: Predicting Opportune Moments for In-vehicle Proactive Auditory-verbal Tasks

  • Metadata:
    • author: Auk Kim, Woohyeok Choi, Jungmi Park, Kyeyoon Kim, Uichin Lee
    • title: Interrupting Drivers for Interactions: Predicting Opportune Moments for In-vehicle Proactive Auditory-verbal Tasks
    • year: 20182018

  • Essay:
    • While consumer-grade autonomous driving is still quite far away, the incorporation of voice assistants for secondary tasks into the driving experience is imaginable in the near future. However, as vocal interaction can also negatively impact driving ability, finding the right moments for voice-based interactions is crucial.
    • This paper pioneers this issue by first defining what an interruptible moment is: The driver's safety must not be inhibited, he must be able to perform well on the interaction, and the interaction should feel easy to him. They then collect data on these metrics by conducting case studies both in simulated and real-world driving environments. In the end, they evaluate multiple classification models that aid the identification of these opportune moments for interaction based on vehicle data and manage to achieve satisfactory accuracy in this task.
    • The paper is the first to offer a clear definition of interruptibility in the new context of voice-based assistance while driving, which lays essential groundwork for future works in this area.
    • I also think works like theirs are a vital step towards making the general populous more receptible to technology-based assistants. As recent trends have shown, one of the most crucial factors when it comes to the adoption of these technologies is the friction of use (or lack thereof). This paper demonstrates a prime example of how this reduction in friction could be approached (identifying fitting moments). Finally, the idea of pervasive computing in traffic (cars communicating with traffic controllers and each other) could offer new data-sources that could further aid the identification of such opportune moments.
    • When discussing their choice of the secondary task for the experiment, they opted for n-back. They dismissed ospan-tasks on account of them being cumbersome to customize for cognitive demand. They cite no source for this, and I am very skeptical of that claim. Having worked with a psychology research team focusing in part on the topic of adaptive ospan-tasks for the last year, I found there are many ways to influence the difficulty of ospan-tasks. Limiting the range and complexity of calculations, varying break times, and increasing/decreasing the number of letters to remember offer a high degree of customizability and should have been a consideration when evaluating ospan for a secondary task.
    • I also found the form of this paper to be slightly below the standard present in the previous papers I have read for this course. Some grammatical errors, like missing articles and incomplete sentences, are spread throughout the text, and a lot of facts (e.g., their measures of interruptibility) are repeated more often than necessary.
    • Lastly, I think that capturing more global features for their classification models (e.g., previous behavior instead of only the vehicle data from a few seconds) could help identify the overall driving situation and improve the accuracy of their models.
    • Despite these minor shortcomings, I overall enjoyed working through this paper (also considering that this was the first paper I read that utilized a videogame - EuroTruckSimulator2 - for data collection).