Application of Collision Probability Estimation to Calibration of Advanced Driver Assistance Systems 2019-01-1133
Until recently, automotive safety technology has been focused on protecting the vehicle and occupants passively in the event of a crash. After rapid development in passive safety, the focus has shifted to mitigate the collision or crash itself with the aid of Advanced Driver-Assistance Systems (ADAS). Therefore, these driver-assistance technologies, especially the ones that compensate for the drivers’ errors are being developed at a blistering pace. In the automotive industry, ADAS are designed and calibrated rigorously to provide robustness against highly uncertain environments in which these systems usually have to operate. Typical calibration procedures for such systems rely extensively on on-road testing in a controlled environment. These procedures are time-consuming, expensive, and often not able to cover all the critical test scenarios that could be encountered by ADAS in the real world. Therefore, simulation-based calibration has been gaining more prominence and emphasis for ensuring high coverage along with easier scalability and usage. This paper attempts to provide an alternative approach for calibrating ADAS in the controller development phase by the aid of simulated test case scenarios. In this study, we execute characterization of the uncertainty in the position and heading of the ego and the obstacle vehicles. This exercise captures the errors in the detection of vehicles’ states in the environment and localization errors of the ego vehicle’s states. Following it, the approach estimates the probability of collision between the two vehicles for a trajectory computed by an ADAS through a Monte Carlo approach. The ADAS intrinsic parameters are then calibrated to meet the collision probability requirements. For illustration purpose, the method is applied on tuning a Lane Change Assistance System for a four-wheel sedan equipped with Short-Range and Long-Range Radar Sensors.