Abstract:
Microscopic car-following models can be applied in Autonomous Vehicles (AVs) to control the real-time longitudinal interactions among individual vehicles. Besides, car- following models can have a vital role in Advanced Vehicle Control and Safety Systems such as collision warning, adaptive cruise control, lane guidance driver assistance, and brake assist, as well as in modeling simulation of safety studies and capacity analysis in transportation science. The car-following models rely on sensor-measured values. How- ever, in reality, sensor measurements are generally inaccurate. Surprisingly, till now to the best of our knowledge, no assessment of the car-following models in the presence of sensor measurement errors for AVs exists. To fill up this gap in the literature, in this thesis, we assess nine prominent car-following models for AVs in the presence of sensor measurement errors in terms of safety, trip times, and flow and fuel efficiency through rigorous simulations covering a real highway map. We show that sensor errors significantly and negatively impact safety and flow in all models, while they do degrade transport efficiency (increase trip times and fuel consumption of some of the models). Moreover, an important finding of our study is that none of the models is highly fault-tolerant and suitable for AVs in the presence of sensor measurement errors. This happens as some models produce collisions and/or negative velocity while all models violate traffic lights in the presence of sensor measurement errors. Nonetheless, we find that the k-leader Fuel-efficient Traffic Model (kFTM) is the most fault-tolerant and the most collision-free model, having reasonable trip times and fuel consumption among our investigated models.
Additionally, to the best of our knowledge, the car-following models available to date are yet to realize and accommodate the impacts of the sensor measurement errors. Therefore, in this thesis, we propose a new fault-tolerant car-following model by realizing and accommodating the sensor measurement errors. We evaluate the proposed fault-tolerant car-following model in the presence of sensor errors based on safety and transport efficiency through rigorous simulations covering a couple of real highway maps. The proposed model improves the level of safety by 97%. However, it cannot nullify the number of collisions.
Therefore, we further propose three model-agnostic strategies to escalate the fault tolerance levels of the car-following models. We evaluate the proposed strategies in the presence of sensor errors based on safety and transport efficiency through extensive simu- lations covering a couple of real highway maps. Our simulation results demonstrate that the proposed strategies can greatly reduce (or even nullify in most cases) the number of collisions that occur for different car-following models in the presence of sensor errors at the cost of minimal degradation in transport efficiency.
