










【摘要】為評估中國高速公路擁堵工況下自動駕駛汽車在切入場景中的安全風險,從自然駕駛數據集中提取64個切入樣本,采用六層次模型和相關性分析方法確定場景要素范圍,通過抽樣生成1 000個測試用例并構建安全評估指標體系分析車輛運行安全,運用隨機森林算法確定引發風險的關鍵因素。結果表明:在1 000個測試用例中,風險場景占比5.3%,縱向相對速度是導致風險的關鍵要素;擁堵工況下,環境車輛速度低于自動駕駛車輛速度23%時形成高風險切入場景,該指標可作為擁堵切入場景下自動駕駛汽車識別風險的預測指標,亦可用于該場景下的事故責任認定。
主題詞:高速公路 擁堵工況 自動駕駛汽車 切入場景 運行安全 隨機森林
中圖分類號:U463.6;U492.8" 文獻標志碼:A" "DOI: 10.19620/j.cnki.1000-3703.20240024
Construction Test Set and Risk Assessment of Cut-in Scenarios for Autonomous Vehicles under Highway Congestion Conditions
Shi Shuaikun1, Zhao Dan1, Ma Mingyue2, Miao Zelin2, Zhou Xiaoji3
(1. People’s Public Security University of China, Beijing 100038; 2. Research Institute for Road Safety of the Ministry of Public Security, Beijing 100062; 3. China Automotive Engineering Research Institute Co., Ltd., Chongqing 401122)
【Abstract】To assess the safety risks of autonomous vehicles during cut-in scenarios on congested Chinese highways, 64 cut-in samples were extracted from a natural driving dataset. Employing a six-level model and correlation analysis, the static and dynamic factors of the scenarios were defined. Subsequently, 1 000 test cases were randomly generated through sampling, and a safety assessment index system was established to analyze the safety of vehicle operations. Lastly, the random forest algorithm was applied to identify the key factors triggering risks. Results indicate that risk scenarios account for 5.3% of the total, with longitudinal relative velocity identified as the crucial factor. Under congested conditions, a high-risk cut-in scenario is formed when the speed of surrounding vehicles is 23% lower than that of autonomous vehicles, this indicator serves as a crucial predictive measure for identifying collision risks in congested cut-in scenarios for autonomous vehicles and may be applied in determining liability of accident in such scenarios.
Key words: Highway, Congested conditions, Autonomous vehicles, Cut-in scenario, Operational safety, Random forest
1 前言
自動駕駛汽車行駛過程中,相較于跟車等場景,切入場景被視為高風險場景之一[1-3]。在切入場景中,擁堵工況下風險更為顯著[4-5]。根據UN R157《裝備自動車道保持系統的車輛核準的統一條款》對高速公路擁堵工況的定義,當平均車速低于60 km/h時,通常可以認為道路處于擁堵狀態,自動駕駛汽車需要更加謹慎地處理切入場景,以確保行駛安全[6]。
目前,場景組合方法[7]作為國內外常用的場景構建方法,更關注基于本體論結構化描述場景和組合關鍵要素。在場景結構化描述方面:Geyer等[8]在控制模塊中引入本體論進行場景結構化表達;Ulbrich等[9]提出知識庫模型,層次化分類場景要素;Menzel等[10]根據要素類型,使用5個獨立分層實現了場景描述;……