An Investigation about the Human Braking Control during the Car-Following Scenarios

在跟車情況下人類制動控制的研究調查

戴志奇(Chih-Chi Dai) ； 陳亮光(Liang-Kuang Chen)

《中國機械工程學刊》 37卷3期 (2016/06) Pp. 239-249

摘要

先進駕駛輔助系統的設計中駕駛人員的控制行為和意圖是至關重要的。在本研究中，針對在跟車情況下的人類制動控制進行研究。而駕駛人的行為測試數據是從車輛配備的感測器所收集。實際駕駛制動資料以K-means技術進行訊號分群之分析、以及使用灰關聯分析來進行煞車事件與行車縱向控制相關之訊號與資訊間的關聯。分群分析之結果顯示，與煞車事件相關之縱向控制訊號與各指標，可將訊號與指標適當地區分成較之小群組，可改善之後的分析與控制計算負擔。K-means的結果同時顯示部分指標在部分較小數量的分群上其Silhouette 係數較低因此應於分析時避免。灰色關聯分析其結果顯示，駕駛在實際制動動作之前，沒有任何單一訊號特徵可以顯著指出駕駛將要踩煞車的行為，這種分析結果可能是由於駕駛者煞車行為之觸發前大多會透過預視與防範式的策略補償，因此煞車動作的觸發被抽象地建立在駕駛者之大腦裡，故不易透過外部駕駛操作的訊號分析而預測。在針對不同煞車事件間煞車期間與不同指標間的關聯分析中可看出，煞車期間的大小與行車時的跟車距離關聯度最高，同時若能透過適當的情境分類可稍微提升煞車期間與各指標間的關聯度。如此可提供根據跟車行為預估後續煞車特徵的有效參考。

並列摘要

The control and intention of the human driver are crucial to the design of advanced driver assist systems. In this research, the human braking control during the car-following scenarios is investigated. The driving data is collected from a test vehicle equipped with the necessary sensors. The data is analyzed to investigate the signal clustering and correlations with actual braking actions, both for the signal variations in the time domain and the metrics variations among different brake events. The k-means technique is employed to determine the suitable number for data clustering, and the results indicate that several of the longitudinal control related signals can be appropriately classified into small clusters to facilitate the analyses and control computations. The k-means analysis with variations among brake events similarly indicates that low number of clustering is suitable, with some exceptions of certain choices of number for certain metrics. The grey relational analyses results indicate that the signals prior to the actual braking actions do not reveal significant correlations with the forth-coming braking actions, probably due to the fact that the human drivers' pre-cautious control leveled off the signal characteristics. The need for the braking actions is abstractly built up inside the human mind and cannot be seen from the external signals. The grey relational analyses also show that the correlations between the brake duration and certain longitudinal control related metrics are higher, and dividing the data sets into smaller groups by properly choosing the criterion can improve the grey relational coefficients. This can provide useful guidelines in predicting braking action characteristics based on the car-following patterns.

並列關鍵字

Data clustering ； grey relation analysis ； human braking behavior