Keep it Real: Investigating Driver-Cyclist Interaction in Real-World Traffic

Title of the Paper

Keep it Real: Investigating Driver-Cyclist Interaction in Real-World Traffic

Paper Information

• Field: Human-Computer Interaction (HCI), Traffic Safety and Interaction Design in Autonomous Driving Technology
• Keywords: Cyclists, Vulnerable Road Users, Autonomous Vehicle-Cyclist Interaction, Observations, Field Study, Naturalistic Study, Eye-Tracking

Research Background and Problem

• Existing Problems or Challenges:

  1. Potential conflicts between cyclists and drivers in shared traffic scenarios are often resolved through social interactions (e.g., eye contact, hand signals).
  2. The proliferation of autonomous vehicles (AVs) may lead to the loss of these social signals, increasing complexity and risks in issues like road priority.
  3. Existing research primarily focuses on AV-pedestrian interactions, with limited attention to the interaction needs of cyclists, failing to encompass diverse traffic scenarios.

• Importance: Autonomous vehicles need to adapt to complex and dynamic traffic scenarios, especially where cyclists are more vulnerable. This necessitates the design of systems that can interact effectively and safely with cyclists, ensuring their safety in mixed traffic environments.

• Motivation and Related Work:

  • Current studies mainly focus on single-scenario interactions or rely on simulation experiments, lacking systematic research on driver-cyclist interaction behaviors in real traffic environments.
  • Autonomous driving requires a more comprehensive knowledge base, including an understanding of cyclists' dynamic behaviors, to design interfaces that meet their needs.
  • Dynamic and complex traffic scenarios (e.g., lane merging, uncontrolled intersections) pose higher demands on the interaction design between cyclists and vehicles.

Solution

• Proposed Methods and Solutions: This paper investigates cyclist-driver interaction behaviors in diverse traffic scenarios through two field studies:

  1. Field Observations: Observed 414 daily traffic encounters between drivers and cyclists, analyzing interaction behaviors across five typical traffic scenarios.
  2. Naturalistic Cycling Study: Equipped 12 commuter cyclists with eye-tracking devices to capture their first-person perspectives, analyzing gaze behavior patterns in different traffic scenarios.

• Innovations:

  1. Systematically recorded and compared driver-cyclist interaction behaviors and information flow in real traffic scenarios.
  2. Provided the first scenario-based direct evidence for designing smarter AV-cyclist interaction systems.
  3. Introduced fine-grained "Area of Interest (AOI)" annotation analysis, offering references for future AV scenario simulation designs.

• Implementation Steps:

  1. Selected high-risk scenarios in urban areas where interactions frequently occur (e.g., lane merging, uncontrolled intersections, roundabouts) to collect interaction behavior data.
  2. Pre-screened cyclists using questionnaires and equipped them for naturalistic experiments, recording eye-tracking data and bicycle movement trajectories to analyze their natural decision-making processes.
  3. Encoded and compared social signals (e.g., gestures, head movements) and implicit signals (e.g., acceleration, deceleration) from both parties.

Research Findings

• Specific Findings:

  1. Cyclist-driver interaction behaviors varied significantly across different traffic scenarios, with more complex interactions observed at uncontrolled intersections and lane merging scenarios.
  2. Cyclists relied more on road markings and traffic signals in controlled scenarios, while focusing on vehicle sides or drivers' social cues in uncontrolled scenarios.
  3. Analysis revealed that dynamic and asymmetric scenarios (e.g., bottleneck sections or complex lane merging) significantly impacted cyclists' behavioral stability, highlighting the need for real-time situational awareness.

• Advantages Compared to Existing Solutions:

  1. Provided behavioral data from real traffic scenarios, offering higher ecological validity compared to simulation studies.
  2. Expanded the design space for AV-cyclist interaction systems, such as external projections, enhanced road markings, and environmental awareness interfaces.

• Experimental or Evaluation Results:

  • Multi-scenario data indicated that the frequency and nature of driver-cyclist interactions were guided by the complexity of the scenarios.
  • Analysis of cyclists' gaze behaviors showed a higher demand for diverse display areas when interacting with dynamic vehicles.

• Limitations and Future Directions:

  1. The study was limited to urban environments in a single city (Glasgow); future research should expand to different countries and rural environments to validate the generalizability of the data.
  2. Only one-on-one interactions were covered; future studies should explore multi-party and multi-vehicle interaction scenarios to investigate scalable designs.
  3. More detailed AOI descriptions and long-term recordings are needed to capture micro-level behavior patterns.
  4. Dynamic and multimodal designs (e.g., integrating AR head-mounted devices) may be a key area for future exploration.

Through these studies, this paper provides foundational knowledge and design guidelines for developing AV-cyclist interaction systems, which can be used to navigate complex traffic scenarios and enhance the safety of shared roads.