Autonomous Vehicle Communication Strategies Modeled in Virtual Reality

We sought to better understand how autonomous vehicle (AV) communication strategies impact human road users’ perceptions and behaviors. More specifically, we explored the impact of different external human-machine interface (eHMI) designs on understanding, task load, comfort, trust, acceptance, and reaction time. To accomplish this, we created virtual reality (VR) scenarios where human participants interacted with AVs. Participants experienced biking, driving, and pedestrian simulators and were brought back after initial testing to explore acclimation and learning effects. In terms of perceptions, the presence of an eHMI was the strongest predictor of understanding, comfort, trust, and acceptance outcomes in the statistical models when controlling for all other variables. There was a clear divide between text-based eHMIs and non-text eHMIs, with text-based eHMIs reporting better perception scores and the LED Windshield reporting the worst perception scores. There were perception acclimation effects detected (most notable for task load and comfort), but they had less of an impact than the presence of an eHMI. Perception outcomes had weaker relationships with participant characteristics than with AV characteristics. While behavioral outcomes should be interpreted with caution because of low participant sample sizes, behavioral results largely mirrored perception results in that significant reductions in reaction time were observed with the presence of an eHMI (3.69 second reduction), yielding (3.16 second reduction), and acclimation (0.134 second reduction per trial). Results suggest that eHMI design, AV behavior, and acclimation are most impactful in terms of both perceptions and reaction time.本研究旨在更深入地探究自动驾驶汽车（autonomous vehicle, AV）的通信策略对人类道路使用者感知与行为的影响。具体而言，本研究考察了不同外部人机交互界面（external human-machine interface, eHMI）设计对理解程度、任务负荷、舒适度、信任度、接受度及反应时的影响。为实现上述研究目标，我们构建了虚拟现实（virtual reality, VR）场景，使人类被试与自动驾驶汽车开展交互。被试依次体验了骑行、驾驶与行人模拟装置，并在初始测试后重返实验，以探究适应效应与学习效应。在感知维度层面，当控制所有其他变量时，外部人机交互界面的有无是统计模型中影响理解程度、舒适度、信任度与接受度结果的最强预测因子。基于文本的外部人机交互界面与非文本外部人机交互界面之间存在显著分界：基于文本的外部人机交互界面获得了更优的感知评分，而LED挡风玻璃式外部人机交互界面的感知评分最低。研究还检测到了感知适应效应（在任务负荷与舒适度维度上表现最为显著），但其影响程度弱于外部人机交互界面的有无。感知结果与被试个体特征的相关性，弱于其与自动驾驶汽车特征的相关性。由于被试样本量较小，行为结果的解读需谨慎；但整体而言，行为结果与感知结果大体一致：配备外部人机交互界面可使反应时显著缩短（缩短3.69秒），避让场景下的反应时缩短3.16秒，而适应效应则使每轮测试的反应时缩短0.134秒。研究结果表明，外部人机交互界面设计、自动驾驶汽车行为以及适应效应，均对感知结果与反应时产生显著影响。