駕駛人模擬器實驗結果討論 - 應用車輛駕駛機率模型發展整合型車輛安全控制系統

年度一結果：

實驗 I：駕駛人轉向行為實驗結果

從實驗得知，駕駛人模擬器在在彎道行駛時，駕駛人對於翻覆及側滑現象較為重視，

亦與一般駕駛人過彎形為類似。使車輛不至於發生事故，因此分析駕駛人與安全系統互動行為時，將 β 及

LTR

列入駕駛人之 payoff function 考量為合理的選擇。另外，若僅是使

用相關車輛參數誤差之最大值作為建立駕駛人 payoff function 之基礎仍有其不足性，需考慮各變數之整體響應情形。然而，駕駛人的行為難以預測，因此本研究為了兼顧暫態及穩態響應，因此本研究所使用 ISE 作為駕駛人之 payoff function 建立基礎為合理的選擇。

實驗 II：駕駛人煞車行為實驗結果

此部分同樣使用 ISE 方式來計算煞車量，原因在於駕駛人並非一直踩煞車，因此當駕駛人煞車時，選用 ISE 方式可計算駕駛人之輸入煞車量，而不煞車時 ISE 值亦不會持續增加，因此選用 ISE 做為輸入煞車輛之評估依據。

由實驗結果，目前僅能知道駕駛人會因預期 ESP 介入而減少其輸入煞車量，但減少之比例仍未有合適的評估標準，而當 ESP 介入操控時，有可能會使得車輛在翻覆上的表現劣化。

年度二結果：

實驗 I：卡爾曼濾波器整合線上駕駛狀態判別系統進行驗證效能

本項實驗規劃實際駕駛人在 (A)精神狀態良好(B)精神狀態不佳兩種不同條件下進行操作駕駛模擬器實驗，藉此驗證類神經網路狀態判別系統與 DSP 實現的狀態判別系統透過 卡爾曼濾波器整合的效能改善實驗。

【Case A】駕駛精神狀態良好的實驗

圖 6-3 卡爾曼濾波器整合駕駛疲勞指標

【Case B】駕駛精神狀態不佳的實驗

圖 6-4 卡爾曼濾波器整合駕駛疲勞指標

由實驗 Case A 與 Case B 結果可以知道，本研究所建立的卡爾曼疲勞濾波器整合 PNN 與透過 DSP 所建的兩套駕駛狀態判別系統確實能有效改善整體系統的準確度與可靠度，讓系統較不會受到外在干擾或模型正確性降低的影響而導致駕駛人狀態判別錯誤。

實驗 II：模糊邏輯識別「Driver Override 行為與否」之效能驗證

本項實驗規劃實際駕駛人在低、中和高疲勞三種不同狀態來進行模擬器實驗，並在指定時間內要求駕駛人執行特定行為，其中在中和高疲勞狀態包含兩種不同規劃實驗分別為 (A)有加入障礙物(B)無加入障礙物。藉由這些實驗以比較有加入駕駛人資訊及未加入駕駛人資訊的模糊邏輯識別出 Driver Override 之效能優劣。

實驗結果顯示出出將駕駛人資訊列入識別駕駛介入主動轉向輔助系統與否的操作行為的考量可提高正確的識別率，並且可以看出有考慮駕駛人資訊的

β

_r權重值切換時間會比未考慮駕駛人資訊來的延後，這將會使駕駛減鍰車輛變換到另一車道的速度而不影響到駕駛的操控行為，即為了保護此時精神狀況稍差的駕駛人。

圖 6-5 各駕駛在中疲勞狀態之無障礙物實驗數據分析卡爾曼濾波器整合駕駛疲勞指標

圖 6-6 各駕駛在高疲勞狀態之無障礙物實驗數據分析卡爾曼濾波器整合駕駛疲勞指標

七、結論及未來工作

年度一中提出以改善率的觀點評估安全系統啟動後對於車輛動態表現的影響，此方式仍有其不足之處，係因若是安全系統啟動後反而使得某種車輛表現劣化，則使用改善率的方法則非一良好選擇

而在年度二中，我們改以 DSP 實現駕駛眼睛狀態的識別機制而再經由離線的動態影像分析結果顯示有一定的吻合度，並且藉由此資訊建立出一線上駕駛疲勞指標系統。接著再探討兩套不同特性基礎的駕駛狀態判別系統，分別由影像處理演算法以及機率類神經網路所建構，再透過卡爾曼濾波器整合這兩套獨立的線上駕駛資訊，從實驗結果顯示出經整合後可確實改善整體系統的準確度與可靠度，讓系統較不會受到外在干擾或模型正確性降低的影響而導致駕駛人狀態判別錯誤。然後，在透過模糊邏輯識別出 DO 的駕駛意圖

在駕駛意圖的判別，可嘗試其他識別方法，例如機率類神經網路 (Probabilistic Neural Network, PNN)、隱馬爾可夫模型 (Hidden Markov Models, HMMs)、支持向量機 (Support Vector Machine, SVM)…等。而我們將在未來嘗試使用隱馬爾可夫模型來做為駕駛意圖的判別，希望能更準確的判別出駕駛之意圖。

八、參考文獻

[1] Yi, K., Chung, T., Kim, J. and Yi, S., “An Investigation Into Differential Braking Strategies for Vehicle Stability Control.” Proceedings of the Institution of Mechanical Engineers;

2003; 217, 12; ProQuest Science Journals p1081-1093

[2] Li, J., Xu, B., Zhang, Y., Liu, W. and Liu, M., “Control Algorithm Based on Neural Network PID Controller For Vehicle Electronic Stability Program.” Jilin Daxue Xuebao

(Gongxueban)/Journal of Jilin University (Engineering and Technology Edition), v 37, n 4,

July, 2007, p 741-744

[3] Goodarzi, A. and Esmilzadeh, E., “An Optimal Vehicle Stability Enhancement Strategy For Articulated Vehicle.” Proceedings of 2006 ASME International Mechanical Engineering

Congress and Exposition, IMECE2006 - Design Engineering, 2006, p9

[4] Ungoren, A.Y. and Peng, H., “An Adaptive Lateral Preview Driver Model,” Vehicle System

Dynamics, v43, n4, April, 2005, p245-259.

[5] Chen, B.C. and Peng, H. “Differential-braking Based Rollover Prevention for Utility Vehicle with Human-in-the-loop Evaluation”, Vehicle System Dynamics, v 36, n4-5, 2001, p 359-389

[6] Lu, J., Messih, D., Salib, A. and Harmison, D., “An Enhancement to Electronic Stability Control System to Include a Rollover Control Function.” SAE 2007-08-0809。

[7] Johansson, B. and Gafvert, M., “Untripped SUV rollover detection and prevention,” IEEE

Conference on Decision and Control (CDC), Atlantis, Paradise Island, Bahamas, 2004, p

5461-5466.

[8] Paul, J. Th., “Reference manual for a lane keeping simulation tool,” The University of

Michigan Transportation Research Institute, July 1995.

[9] Wierwille, W. W., Ellsworth, L. A., Wreggit, S.S., Fatrbanks, R.J., and Kirn C.L., “Research on vehicle-based driver status/performance monitoring; development, validation, and refinement of algorithms for detection of driver drowsiness,” National Highway Traffic

Safety Administration Final Report: DOT HS 808 247 , 1994.

[10] Dinges, D. F. and Grace, R., ”PERCLOS : a valid psychophysiological measure of alertness as assessed by psychomotor vigilance,” No. FHWA-MCRT-98 006.Federal Highway

Administration Tech Brief. Publication., 1998.

[11] Grace, R., ”Drowsy driver monitor and warning system,” Pittsburgh, Robotics Institute

Carnegie Mellon University, 2001.

[12] 孫宗瀛,楊英魁“Fuzzy 控制：理論、實作與應用” 全華科技圖書股份有限公司, 台北 , 2001.

[13] 彭孟璿,“線上駕駛人建模與駕駛狀態判別之正確性評估,”

國立臺灣科技大學機械工

程系碩士論文

, 台灣台北, 2007.

出席國際學術會議心得報告

計畫編號 NSC 96－2221－E－011－131－MY2

計畫名稱應用車輛駕駛機率模型發展整合型車輛安全控制系統出國人員姓名

服務機關及職稱陳亮光臺灣科技大學機械系助理教授

會議時間地點 Kempinski Hotel, Xi'an, Shaanxi, China, June 3-8, 2009 會議名稱 2009 IEEE Intelligent Vehicle Symposium

發表論文題目 Control Authority Determination of a Vehicle Lane Keeping Assist Controller 三、參加會議經過

本次會議由 IEEE Intelligent Transportation System 協會主辦，除了三個 keynote speeches 外，本研討會的特色在於所有口頭演講的文章報告均在 single track oral

presentation 進行，由於參與者均是對智慧型運輸系統有興趣之人士，同時研討會之主題定義明確且不廣，因此這個做法有其背景與原因。搭配 Prof. Broggi 與 Prof. Ozguner 的演講以及少數幾篇的口頭報告，此次研討會展示了許多的論文海報。每位海報投稿者須在展出時間在所展示的海報前與所有的與會人士溝通並討論，出差人同時受邀擔任該次展出時段之 co-chair。Prof. Werbos 主講有關如何避免使用石油能源的演講則搭配著研討會晚宴舉行。會議最後一天同時還參觀了在西安交通大學主辦的中國無人駕駛車的競賽。

四、與會心得

雖然研討會主題為智慧型車輛，但參與會議之文章主題涵蓋甚廣，內容遍及車輛主動安全、運輸、人機介面、以及駕駛人模型等。透過此次會中報告與世界各國研究學者的討論，對本計畫之未來研究方向與內容有很好的幫助。由於在大陸舉行，也看到了許多大陸各學校所報告的成果，很明顯的大陸在車輛產業上投注了大量的資源。會中也可感受到世界各國再車輛技術上目前的一個重點就是與駕駛輔助相關的主動式安全系統。

不管是警示或是動態控制，都有許多團隊與單位在進行研究。期望能透過此次的會議互動，使本計畫之研究能持續修正並與國際接軌。

Abstract—The vehicle lane keeping assist through active front wheel steering is developed in this research using robust model reference adaptive control. Two independent on-line driver drowsy assessment systems are employed to determine the driver drowsy indices. The two sets of drowsy assessments will be integrated through a Kalman filter to help determine the control authority of the controller. Computer simulations show that the proposed controller and control authority scheduling algorithms achieve the desired functions. Driving simulator experiments will be conducted to verify the proposed framework.

I. INTRODUCTION

OST vehicle crashes are caused by human mistake, and significant effort has been devoted to develop active safety system to help prevent vehicle accidents. The active safety systems refer to the active devices that function prior to the incidences of the crashes with the purpose to help avoid the crashes. A crucial characteristic of these systems is that they generally function together with the human driver. The interactions between the human drivers and the active safety systems affect the performance of the overall vehicle safety.

Consequently, a careful consideration of the human drivers in the design of active safety systems is imperative for the success of their designed functions. The human driver behavior has been extensively investigated in the literature, ranging from the human factor research to the control oriented driver models, e.g., [1-8]. However, the applications of the existing driver models to the development of active safety systems are rare. The main reason is that most driver models are developed to mimic an averaged driver behavior based on the data collected from human driving. While the models present meaningful characteristics of driver behavior, they lack the timely nature to represent the variation in driver behavior and may be too complex to estimate in real time.

To serve as a useful resource for the operations of the vehicle active safety systems, attempts to use low order driver model structures and estimate their parameters in real-time has been reported in the recent literature, e.g., [9]. Granted

Manuscript received January 14, 2009. This work was supported in part by the National Science Council of Taiwan under Grant NSC 96-2221-E-011-131-MY2

Liang-kuang Chen and Chuan-hui Yang are with the National Taiwan University of Science and Technology, Taipei, Taiwan (phone:

+886-2-2737-6481; fax: +886-2-2737-6460; e-mail:

lkchen@mail.ntust.edu.tw).

that the estimated driver models are not exact and can be absurd sometimes, it is possible to obtain an estimation of the driver model uncertainty level from the vehicle driving data, as reported in [10]. Using the on-line estimated driver models, together with the model uncertainty level, the active safety systems can be modified to be more driver-oriented. It is natural to expect that the active safety systems can be tuned to provide more aggressive control actions based on the real-time driver model information, thus improving their functionality and safety contribution. In this research the vehicle lane keeping assist system is selected as the application target to investigate.

An on-line driver modeling and state assessment previously developed is employed to provide the real-time driver information [9]. As illustrated in Fig. 1, an adaptive lane keeping assist controller using active front wheel steering is designed to compensate the difference between the actual driver behavior and an “ideal” driver behavior. The adaptive controller is a model reference adaptive control (MRAC) such that the compensated driver behavior tracks the reference driver model which has been tuned a priori to be a high performance driver model. If the driver model variation is significantly slower that the adaptation, the adaptive rule can be designed using Lyapunov stability criterion and the model matching error will converge eventually. In effect the MRAC is a parallel co-pilot trying to achieve the automated steering control. The driver’s control actions can be regarded as disturbances within the control authority of the steering assist controller. A natural question arises in the determination of the control authority. Due to the variation in driver state and preferences, the control authority is generally determined based on human acceptance with repeated driving experiments, as reported in [11] for example.

Furthermore, significant modeling error in the control design causes serious robustness problem in the overall system performance. In this research, robust MRAC is developed to provide the steering assist control and a novel approach to tune the control authority on-line based on the instant driver information is proposed. The driver acceptance of this design will be evaluated using PC based driving simulator.

Control Authority Determination of a Vehicle Lane Keeping Assist Controller

Liang-kuang Chen and Chuan-hui Yang Department of Mechanical Engineering

National Taiwan University of Science and Technology

M

Fig. 1. Block diagram of the lane keeping steering assist controller

II. RMRAC STEERING ASSIST CONTROL

在文檔中應用車輛駕駛機率模型發展整合型車輛安全控制系統 (頁 20-31)