Development and Evaluation of Deceleration Intent Inference System by Unscented Kalman Filter
Development and Evaluation of Deceleration Intent Inference System by Unscented Kalman Filter
复制标题
无迹卡尔曼滤波器减速意图推断系统的开发与评估
DOI：
10.11351/jsaeronbun.52.343
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
若林翔，鈴木宏典
中科院分区：
文献类型：
--
作者：
Ping Chi Yuen;Kenji Sasa;Hideo Kawahara;Chen Chen;川原 秀夫,川上 拓也,笹健児;若林翔，鈴木宏典
即使车辆配备了高度发达的自动驾驶系统，工作刹车灯也只是识别前车减速的方式。提前识别减速意图将降低追尾风险，尤其是在高密度、高速跟车状态下。作者一直在开发一种系统，可以在驾驶员最有可能采取的行动之前 1.5 秒推断出减速意图。然而，之前使用无迹卡尔曼滤波器（UKF）的系统由两个独立的过程组成，即车辆队列状态估计和减速意图预测。这种分离会导致对意图的高估或低估，从而降低预测精度。本文旨在通过将两个过程结合在一个过程中来改进意图推理系统。此外，引入人工神经网络模型和多元回归模型来估计车辆状态变量以及推断意图。数值分析表明，与之前的系统相比，修订后的模型为所有五名参与者提供了更准确的意图。可以得出结论，集成状态反馈系统不仅适用于状态估计，而且适用于减速意图推断的预测。
The working brake lights are only the way to recognize the deceleration of preceding vehicle even if the vehicle is equipped with a highly developed autonomous driving system. The prior recognition of deceleration intent would cause a mitigation of a risk of rear-end collision especially in a high-density and high-speed car-following state. The authors have been developing a system that infers deceleration intention 1.5 s in advance of its driver's most likely action. However, the previous system that utilizes an unscented Kalman filter (UKF) consists of two isolated processes, a state estimation of vehicle platooning and a prediction of deceleration intention. This separation causes an over-or under-estimate of the intent to worsen the prediction precision. This paper aims to improve the intent inference system by combining both processes within a single procedure. In addition, artificial neural network model and multiple regression model are introduced to estimate both vehicle state variables together with the inferred intention. Numerical analyses showed that the revised model provided more accurate intention compared to the previous system for all five participants. It can be concluded that the integrated state feedback system is appropriately working not only for the state estimation but also the prediction of deceleration intent inference.