How Self Embodiment and Agent Position Affect Safety-Conscious Behaviors and Performance in Guiding Peer Agents Across Traffic in VR

This study examines how embodiment and safety-conscious behaviors shape strategies and performance when guiding virtual peers across traffic in immersive road-crossing scenarios. Participants physically walked in tracked space, with motion sensors and the HTC VIVE system mapping movements into the virtual environment. They were tasked with leading same-gender peers safely across a one-way road with continuous traffic, requiring vigilance to protect both themselves and the peer. Two between-subject factors were included: self-Avatar presence (Full-Body vs. Iconic) and participant gender (Male vs. Female). Two within-subject factors were manipulated: Agent Position (Trailing vs. Beside) and Vehicle Proximity Relationship (participant closer vs. peer closer to approaching vehicles). Results showed that crossing strategies and safety outcomes were significantly shaped by agent position, vehicle proximity, gender, and avatar presence. Walking side-by-side supported safer, more conservative behaviors with larger time margins and fewer collisions, whereas trailing increased participant difficulty and risk. Vehicle positioning also mattered: peers opposite traffic (AUV) led to hesitation and smaller safety buffers, while same-side positioning (UAV) promoted earlier entry, larger margins, and safer crossings. Avatar presence and gender further influenced strategies: avatars amplified cautious behavior, females generally maintained larger safety margins, and males showed more collisions in trailing conditions with distinct timing adjustments. These findings have implications for VR-based road-crossing training, underscoring the need to account for peer positioning, traffic context, and user characteristics in moderating user behavior and towards designing effective learning environments.