LITERATURE REVIEW ON AUTONOMOUS DRIVING DATASETS AND PARAMETER CALIBRATION FOR CAR FOLLOWING EVENTS

Abstract

Autonomous vehicles (AVs) are expected to reshape future transportation systems. It is crucial to have high-quality datasets for the development of reliable driving algorithms and making important decisions in the transportation sector. Additionally, real-world AV-oriented datasets are essential for understanding the impact of AVs on human-driven vehicles (HDVs), traffic flow, and for developing data-driven car-following models. Parameter setting in these models is essential to simulate real-world driving behavior, as AVs can either enhance or hinder transportation network performance depending on settings like time headway, acceleration, and penetration rates. This study reviewed driving behavioral models for AVs, assessed the availability and characteristics of different open access real-world AV datasets, and calibrated parameters for the Wiedemann 99 and Intelligent Driver Model (IDM) car-following models using the Lyft L5 AV dataset. Findings highlight that each dataset has specific applications, underscoring the importance of understanding their characteristics for appropriate usage. Calibration results indicated variability in parameters for various car-following events, necessitating calibration for varied leader-follower scenarios. The Wiedemann 99 model showed that AVs following HDVs maintain a smaller time headway (CC1) compared to HDVs following AVs, whereas the IDM showed the opposite. Also, AVs tend to maintain longer spacing and time gap (CC4 and CC5) while following HDVs, and take less time (CC3) to decelerate and maintain a safe distance, compared to HDV-following-AV cases, which also similar to IDM based results (s₀). Notably, significant behavioral changes in human drivers occur when interacting with both AVs and other HDVs in mixed-traffic conditions.