首页 >> 学术论文 >> 自动驾驶安全

Calibration and Evaluation of the Responsibility-Sensitive Safety Model of Autonomous Car-Following Maneuvers Using Naturalistic Driving Study Data

发表时间:   阅读次数:

Abstract: Safety guarantees are vital to the dependability of the automated vehicle (AV), so are of primary concern to the AV industry and regulatory bodies. Responsibility-Sensitive Safety (RSS), proposed by Mobileye, is a rigorous mathematical model that defines the real-time safety distance that the AV needs to maintain in relation to surrounding vehicles, and helps the AV respond to dangerous situations. However, RSS’s performance has not yet been tested with real driving data. This study thus calibrates and evaluates the RSS model based on car-following scenarios created from safety-critical events (SCEs) detected in the Shanghai Naturalistic Driving Study. SCEs were identified from the raw data with a trigger method; they were manually validated and then converted into virtual scenarios in which the human-driven subject vehicle was replaced with an AV controlled by a model predictive control based adaptive cruise control (ACC) system embedded with RSS. The RSS model was calibrated with the goal of achieving optimal safety performance, that is, of generating the lowest time-integrated time-to-collision (TIT) after the AV algorithms were implemented in the safety-critical scenarios. By comparing the performance of RSS-embedded ACC, ACC-only, and the unassisted human driver, this study found that: 1) RSS-embedded ACC performed best in eliminating underlying safety-critical situations by controlling the average TIT for each 15-second event at 2.65 s 2 ; 2) RSS-embedded ACC generated the highest average relative speed and relative distance, and the lowest standard deviation of speed; and 3) on average, RSS-embedded ACC was able to perceive the SCEs 2.31 s ahead of the human driver.

 

Xiaoyan Xu, Xuesong Wang, Xiangbin Wu, Omar Hassanin, Chen Chai. Calibration and Evaluation of the Responsibility-Sensitive Safety Model of Autonomous Car-Following Maneuvers Using Naturalistic Driving Study Data. Transportation Research Part C: Emerging Technologies, 2021, 123: 102988.

©CopyRight 2003-2012   同济大学交通运输工程学院

备案号:沪ICP备13005359号-1