[Paper Review] Gentlemen on the Road: Understanding How Pedestrians Interpret Yielding Behavior of Autonomous Vehicles using Machine Learning

Yoon Kyung Lee, Yong-Eun Rhee+2 more|arXiv (Cornell University)|May 16, 2020

Autonomous Vehicle Technology and Safety41 references4 citations

TL;DR

This study investigates how pedestrian trust and crossing behavior are influenced by autonomous vehicle (AV) yielding behavior and size using a virtual reality experiment with 39 participants. By applying machine learning to classify head orientation patterns, the authors find that pedestrians are most attentive to non-yielding large AVs, and both yielding behavior and vehicle size significantly affect perceived safety and crossing decisions.

ABSTRACT

Autonomous vehicles (AVs) can prevent collisions by understanding pedestrian intention. We conducted a virtual reality experiment with 39 participants and measured crossing times (seconds) and head orientation (yaw degrees). We manipulated AV yielding behavior (no-yield, slow-yield, and fast-yield) and the AV size (small, medium, and large). Using machine learning approach, we classified head orientation change of pedestrians by time into 6 clusters of patterns. Results indicate that pedestrian head orientation change was influenced by AV yielding behavior as well as the size of the AV. Participants fixated on the front most of the time even when the car approached near. Participants changed head orientation most frequently when a large size AV did not yield (no-yield). In post-experiment interviews, participants reported that yielding behavior and size affected their decision to cross and perceived safety. For autonomous vehicles to be perceived more safe and trustful, vehicle-specific factors such as size and yielding behavior should be considered in the designing process.

Motivation & Objective

• To understand how pedestrians interpret autonomous vehicle yielding behavior in real-time interactions.
• To examine the influence of AV size (small, medium, large) on pedestrian perception and behavior.
• To identify patterns in pedestrian attention through head orientation tracking during AV encounters.
• To explore the relationship between perceived safety, yielding behavior, and crossing decisions using qualitative and quantitative data.
• To inform AV design by identifying vehicle-specific factors that enhance trust and safety perception.

Proposed method

• Conducted a controlled virtual reality experiment with 39 participants exposed to AVs exhibiting three yielding behaviors: no-yield, slow-yield, and fast-yield.
• Varied AV size across three levels: small, medium, and large, creating a 3×3 experimental design (size × yielding behavior).
• Collected real-time data on crossing times (in seconds) and head orientation (yaw degrees) during AV approaches.
• Applied machine learning to cluster head orientation changes into six distinct behavioral patterns over time.
• Used post-experiment interviews to gather qualitative insights on perceived safety and crossing decisions.
• Combined quantitative behavioral data with qualitative feedback to analyze the impact of AV characteristics on pedestrian behavior.

Experimental results

Research questions

• RQ1 How does the yielding behavior of autonomous vehicles (no-yield, slow-yield, fast-yield) influence pedestrian crossing decisions and attention patterns?
• RQ2 To what extent does the physical size of an autonomous vehicle (small, medium, large) affect pedestrian perception of safety and willingness to cross?
• RQ3 What patterns of head orientation change do pedestrians exhibit when approaching autonomous vehicles with different yielding behaviors and sizes?
• RQ4 How do pedestrians' subjective perceptions of safety and trust correlate with objective behavioral metrics like crossing time and head movement?
• RQ5 What vehicle-specific design factors (size and yielding behavior) most significantly influence pedestrian trust in autonomous vehicles?

Key findings

• Pedestrians fixated on the front of the AV for the majority of the time, even when the vehicle was approaching closely.
• The highest frequency of head orientation changes occurred when a large AV did not yield, indicating heightened attention and potential threat perception.
• Participants reported that both yielding behavior and AV size significantly influenced their decision to cross and their perception of safety.
• Machine learning successfully classified head orientation changes into six distinct temporal clusters, revealing dynamic attention patterns during AV interactions.
• Fast-yielding AVs were associated with higher perceived safety, while non-yielding large AVs elicited the most cautious and attentive pedestrian responses.
• The combination of vehicle size and yielding behavior had a stronger influence on pedestrian behavior than either factor alone.