Design and Evaluation of Lane-Change Collision Avoidance Systems in Semi-Automated Driving

Abstract

Hands on the steering wheel partial driving automation, in which the system controls the lateral and longitudinal vehicle motion while the driver holds the steering wheel, monitors the roadway, and intervenes when necessary, is an example of shared control driving. To insure mutual safety in shared control driving, the system also needs to guide the driver's interventions to avoid hazardous actions, such as risky lane changes. This study proposes three steering interventions that activate automatically when the system detects a lane change by monitoring the steering wheel angle inputted by the driver and road section and determining that the two vehicles are on a collision course. A driving simulation experiment with 80 drivers was conducted to compare four conditions: 1) no steering intervention; 2) a haptic steering intervention that increases stiffness against steering toward the lane where a collision may occur; 3) an automatic steering intervention that decouples the driver's steering input and autonomously steered the vehicle away from the hazard; and 4) multiple steering interventions that generate different degrees of haptic and automatic steering interventions based on the time headway and time-to-collision between vehicles. Analysis of driving performance and safety under conditions with and without steering interventions indicate that all participants, at some points, initiated lane changes that are likely to result in a crash during partial driving automation. However, the three interventions effectively reduced lane-change collisions compared to the baseline. While the automatic steering intervention avoided all collisions, the multiple steering interventions were determined to provide sufficient safety while being more thoroughly accepted by the drivers than the haptic and automatic interventions separately. These findings have implications for developing adaptive collision avoidance systems considering user preference and driving performance.