Quantitative Evaluation and Modeling of Action-Reaction Cycles in Interactive Human Driving Behaviors

该项目旨在提供一个框架来量化交通中人类驾驶员之间的交互性，通过将交互意识运用到决策算法设计中，使自动驾驶汽车在真实交通中更具创新性和安全性。为此，我们开发了一个可以预测人类驾驶员行为的模块，它允许我们使用统计机器学习来评估一个人类驾驶员的反应对另一个人类驾驶员行为的依赖程度。最后，我们将量化的依赖关系整合到交互感知轨迹预测模型或控制器中，以验证算法在实际交通中的性能。该项目将有利于加拿大自动驾驶汽车和其他相关领域的发展，如加拿大自然科学与工程领域的人工智能和人机交互。这项研究将系统而坚实地为学术界对先进自动驾驶汽车和工业相关产品的人类交互感知行为的理解和表征搭建桥梁。此外，这项研究可以解释人类驾驶员行为的影响，从而使加拿大的日常交通系统更安全，更高效。

This project aims to provide a framework to quantify the interactivity between human drivers in traffic, enabling autonomous vehicles to be more innovative and safer in real traffic by leveraging interaction-awareness into decision-making algorithm design. To this end, we develop a module that can predict the human driver's behavior, which allows us to evaluate how dependent one human driver's reaction is on the other human driver's action using statistic machine learning. Finally, we integrate the quantified dependency into an interaction-aware trajectory prediction model or controller to validate the algorithm performance in real traffic. This project will benefit Canada's development of autonomous vehicles and other related domains such as artificial intelligence and human-robot interaction in Canada's Natural Sciences and Engineering. This research will systematically and solidly bridge an enhanced understanding and characterization of human interaction-aware behaviors toward advanced autonomous cars in academia and related products in industries. Moreover, this research can account for the influence of human driver behavior, thus making Canada's daily transportation systems safer and more efficient.