Enjoy the Ride Consciously with CAWA: Context-Aware Advisory Warnings for Automated Driving

Erfan Pakdamanian, Erzhen Hu, Shili Sheng, Sarit Kraus, Seongkook Heo, Lu Feng

TL;DR

Researchers have developed CAWA (Context-Aware Advisory Warnings), a system that prepares drivers in semi-automated vehicles for takeovers by issuing "gentle nudges" tailored to their current distraction. By matching the warning modality (e.g., a text message for a mobile gamer) to the specific non-driving task, CAWA significantly improves situational awareness and reaction times compared to traditional, one-size-fits-all auditory alerts.

What Problem Does This Paper Address?

In Level 3 (conditionally) automated driving, the car handles most tasks, but the human must be ready to intervene when the system reaches its limits. This creates a dangerous "out-of-the-loop" problem: drivers often immerse themselves in Non-Driving Related Tasks (NDRTs) like reading, gaming, or watching movies.

When a sudden Takeover Request (TOR) occurs, the driver is often startled and lacks the situational awareness to maneuver safely. Current systems typically use uniform auditory beeps, which can be easily missed if the driver is wearing headphones or can cause panic if they are too loud. The authors seek to bridge this gap by providing "advisory warnings"—pre-alerts that gently bring the driver back into the loop before a critical emergency occurs.

Methods

The authors proposed and tested the CAWA framework, which consists of two main steps:

  1. Detection: Identifying the specific NDRT the driver is engaged in.
  2. Selection: Choosing the most effective warning modality based on that context.

To evaluate this, they conducted a user study with 20 participants in a high-fidelity driving simulator (using the CARLA engine). They mapped specific distractions to specific "non-assertive" warning modalities:

  • Gaming on a phone $\rightarrow$ Text message: Since the driver is already looking at their phone, a notification is highly likely to be seen.
  • Watching a movie on a tablet $\rightarrow$ Vibrotactile (haptic): Vibrations in the seat cut through the immersion of a movie.
  • Reading a physical book $\rightarrow$ Speech-based: An auditory "Please pay attention" message works best when eyes are down.
  • Conversing with a passenger $\rightarrow$ Visual HUD: A windshield-projected warning alerts a driver who is already looking around the cabin.

The researchers used state-of-the-art computer vision for real-time eye-tracking and pupil localization, alongside biometric sensors (PPG for heart rate and GSR for skin conductance) to measure stress and cognitive load.

Datasets

The paper does not use a standard external dataset; instead, it generates a primary dataset through a controlled user study.

  • Participants: 20 individuals (14 male, 6 female), aged 18-32.
  • Scenarios: Four urban takeover situations (fallen trees, work zones, police roadblocks, and broken-down cars).
  • Data Collected: 456 total takeover events, including vehicle trajectories, reaction times, eye-gaze vectors, and physiological stress markers.

Key Insights and Ideas

  • The "Modality Match" Hypothesis: The core "aha moment" is that a warning shouldn't just be loud; it should be compatible with the driver's current sensory channel. If a driver is visually occupied by a phone, a visual notification on that same device is more effective than a generic dashboard light.
  • Non-Assertive Priming: Unlike a TOR, which demands immediate action, CAWA's advisory warnings are "non-assertive." They aim to increase "readiness" rather than force a maneuver, reducing the startle effect.
  • Attention Demand vs. Workload: Interestingly, while CAWA improved performance, it didn't significantly lower the overall perceived workload (except for the "attention demand" subscale). This suggests that while the system makes the driver safer, the act of switching tasks remains cognitively taxing.

Key Results

The study compared CAWA against a "baseline" of uniform speech-based warnings:

  • Faster Reaction Times: CAWA led to significantly faster reaction times across all NDRTs ($p < 0.001$).
  • Better Vehicle Control: Drivers using CAWA showed significantly less lateral deviation (lane weaving) during the takeover process.
  • Enhanced Situational Awareness: After receiving a CAWA alert, 87.6% of drivers looked back at the road before the actual TOR was issued. Overall, CAWA increased the likelihood of road monitoring by 26% compared to previous studies.
  • User Preference: 17 out of 20 participants rated CAWA as "gentler" and "safer" than the baseline, though some found the haptic (vibrotactile) modality "shocking" if they weren't expecting it.

Limitations and Future Work

  • Unimodal Constraints: The current study only tested one modality at a time per task. Real-world applications might benefit from multi-modal warnings (e.g., a text and a vibration).
  • Simulator Validity: As with all simulator studies, there is a gap between lab behavior and "wild" driving behavior where real-world consequences exist.
  • Fixed Timing: The advisory warnings were triggered at a fixed interval (~40 seconds before the TOR). Future versions of CAWA could use AI to dynamically trigger warnings based on real-time risk assessment and the driver's level of "deep" immersion.
  • Alert Fatigue: The authors acknowledge that frequent advisory warnings could lead to annoyance or "alert fatigue" if not tuned correctly.

Summary updated

Abstract

In conditionally automated driving, drivers decoupled from driving while immersed in non-driving-related tasks (NDRTs) could potentially either miss the system-initiated takeover request (TOR) or a sudden TOR may startle them. To better prepare drivers for a safer takeover in an emergency, we propose novel context-aware advisory warnings (CAWA) for automated driving to gently inform drivers. This will help them stay vigilant while engaging in NDRTs. The key innovation is that CAWA adapts warning modalities according to the context of NDRTs. We conducted a user study to investigate the effectiveness of CAWA. The study results show that CAWA has statistically significant effects on safer takeover behavior, improved driver situational awareness, less attention demand, and more positive user feedback, compared with uniformly distributed speech-based warnings across all NDRTs.
BibTeX
@inproceedings{pakdamanian2022enjoy,
  author = {Erfan Pakdamanian and Erzhen Hu and Shili Sheng and Sarit Kraus and Seongkook Heo and Lu Feng},
  title = {{Enjoy the Ride Consciously with CAWA: Context-Aware Advisory Warnings for Automated Driving}},
  booktitle = {UI},
  year = {2022},
  doi = {10.1145/3543174.3546835},
}

Loading related content...

Figure