Enhance SIL Simulation Through Driver Behaviour Modeling at Unprotected Left-Turn Scenario for Autonomous Driving SOTIF Analysis

Abstract

As the rapid advancement of artificial intelligence (AI), information and communication technologies, autonomous driving system (ADS) has increased permeation into the traditional automotive industry in recent years. To reduce the Safety of the Intended Functionality (SOTIF) risk of autonomous driving system hence improving its dependability, SIL simulations are extensively exploited as virtual mileage test in compensation of the prohibitively expensive and inefficient road test. In SIL simulation, unprotect left-turn is an intricate traffic scenario to be reproduced due to the intensive interaction between vehicles at the intersection. However, most state-of-the-art commercial simulation software omit the interaction modeling. Thus, in this paper, we proposed a driver behavior modeling approach at unprotected left-turn scenarios to enhance the authenticity of SIL simulation. The left-turn scenario was modelled through three stages, including interaction selection, interaction decision and driver behavior modeling, of which a logit model and intelligent driver model (IDM) were used for the latter two stages. After model calibration, it proves this approach can generate highly authentic traffic flow with unbiased feature distribution towards the real-world, indicating its potential in SIL simulation performance improvement.