Don’t Let the Pigeon Drive the Bus! The Impact of a Bird’s Eye View on Situational Awareness in Remote Driving

Paper Title

Don’t Let the Pigeon Drive the Bus! The Impact of a Bird’s Eye View on Situational Awareness in Remote Driving

Publication Info

• Topic area: Teleoperation interfaces for autonomous vehicles, focusing on situational awareness.
• Keywords: Autonomous vehicles, teleoperation, situational awareness, bird’s eye view, interface design, cognitive load, remote driving, SAGAT, human factors, driving simulation.

Background and Problem

• Problem / challenge: Current teleoperation interfaces for autonomous vehicles (AVs) rely heavily on frontal camera views, which may limit situational awareness (SA). Adding a bird’s eye view (BEV) could theoretically enhance SA but might also increase cognitive load or distract operators.
• Significance: Teleoperation is critical for managing complex scenarios where AVs fail, such as road obstacles or emergencies. Effective interface design is essential to ensure safety and operator performance.
• Motivation and related work: Prior research has explored BEV in driving and robotics, showing potential benefits in spatial awareness. However, its effectiveness in AV teleoperation remains unclear, particularly regarding its impact on SA and cognitive load.

Solution

• Proposed approach: Investigate the impact of incorporating BEV into tele-driving interfaces through two studies: a remote video-based study and a laboratory-based driving simulation.
• Novelty:
  1. Systematic evaluation of BEV integration in teleoperation interfaces.
  2. Examination of BEV’s impact on SA across different interface layouts and driving scenarios.
  3. Analysis of cognitive load and gaze patterns in a driving simulator.
  4. Identification of potential trade-offs between added information and cognitive burden.
• Procedure and key techniques:
  • Study I: Remote video-based experiment with 150 participants comparing three interface layouts (Frontal, Minimized BEV, Gallery BEV) using post-video SA questionnaires.
  • Study II: Laboratory driving simulation with 41 participants comparing two layouts (Frontal, Gallery BEV) using SAGAT-based SA questions, NASA-TLX workload assessments, and gaze tracking.

Results

• Concrete findings:
  • Study I: BEV did not improve SA for nearby or partially visible objects and reduced SA for distant objects and fine details. Frontal-only view outperformed BEV layouts in these scenarios.
  • Study II: No significant differences in SA or cognitive load between Frontal and Gallery BEV conditions. Participants rarely used the BEV display (6% of driving time).
• Advantage over baselines: Frontal-only view demonstrated better performance in detecting distant objects and fine details, with no added cognitive burden compared to BEV layouts.
• Experiments / evaluation:
  • Study I: 142 participants analyzed after exclusions; 8 driving scene videos; 4 SA questions per video; statistical analysis using chi-square tests.
  • Study II: 39 participants analyzed; 8 driving simulations; SAGAT-based SA questions; NASA-TLX workload assessments; gaze tracking for BEV usage.
• Limitations and future work:
  • Study I: Remote setting limited control; focused only on perception-level SA.
  • Study II: Small sample size; limited statistical power for medium/small effects.
  • Both studies: Assumed zero-latency teleoperation; BEV camera height fixed at 80 meters; no BEV-only condition tested.
  • Future work: Explore dynamic BEV configurations, latency effects, alternative driving scenarios (e.g., parking, obstacle navigation), and BEV-only interfaces.

Summary

This study evaluated the impact of adding a bird’s eye view (BEV) to tele-driving interfaces for autonomous vehicles. Across two experiments, BEV integration showed no significant overall benefits for situational awareness (SA) and, in some cases, reduced SA for distant objects and fine details. Cognitive load was not significantly affected, but participants rarely used the BEV display during active driving. These findings suggest that simpler, frontal-only interfaces may be more effective for standard driving tasks, though BEV utility in specialized scenarios warrants further investigation. Future research should explore adaptive BEV configurations, latency effects, and diverse driving contexts to optimize teleoperation interfaces.