Multisensory Contributions to Presence in CARLA-Based Virtual Reality Bicycle Simulation

Engineering notes

Post-hoc pairwise comparisons indicated that, while head-mounted displays (HMDs) provide a level of spatial orientation, the concurrent integration of visual self-embodiment and speed-dependent airflow acts as a unifying sensory cue, significantly increasing system coherence.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Future traffic scenarios will involve autonomous vehicles (AVs) interacting with other road users, including cyclists. For this interaction to be safe and socially acceptable, there is a need for development and testing environments that enable scenarios between AVs and cyclists including ethical, logistical, and safety constraints. While human-in-the-loop transport simulation offers an alternative, achieving high behavioral validity requires maximizing user presence and control. This study evaluates a modular, multi-sensory cycling simulator built on the CARLA autonomous driving engine. Through a fully randomized and counterbalanced within-subjects design with 31 participants, we analyzed four simulator configurations incorporating combinations of stereoscopic visualization, speed-dependent haptic eolic feedback, first-person visual self-embodiment, and 3D spatialized audio. Non-parametric statistical evaluations revealed significant differences across the groups (χ2(3)=60.47, p<0.001, W=0.650). Post-hoc pairwise comparisons indicated that, while head-mounted displays (HMDs) provide a level of spatial orientation, the concurrent integration of visual self-embodiment and speed-dependent airflow acts as a unifying sensory cue, significantly increasing system coherence. Conversely, the addition of 3D spatialised audio yielded no statistically significant improvement (Z=−0.06, p=1.000), suggesting a sensory saturation effect. Cross-dimensional analysis confirmed a positive correlation between perceived presence and interface control (ρ=0.693, p<0.001). These findings suggest that a combination of visual feedback, haptic airflow, and self-embodiment may be enough to provide a realistic environment for cyclist–AV interaction studies including multiple agents.

Links and sources

Comments