Longitudinal Control for Self-driving Cars with Traffic Flow Considerations: Theory, Design, and Experiments

Self-driving cars are around the corner, quite literally. As the industry is spending most efforts on improving safety in corner cases, the impacts of autonomous vehicles (AVs) on traffic congestion are overlooked, possibly due to a lack of regulation, as a result, current adaptive cruise control (ACC) will exacerbate congestion. This dissertation addresses this research gap between self-driving and congestion. It develops new theories, algorithms, and experimental methods for ACC by incorporating traffic flow knowledge. The major findings include the following: i) identification of the research gap that existing datasets and learning methods in the self-driving industry have not well accounted for AVs' impact on traffic congestion, ii) significance of low-level control to string stability under ACC, which has been overlooked so far, iii) incorporation of driver relaxation into commercially-available ACC systems, which proves to be efficient in reducing lane-changing disruptions, iv) a family of novel model predictive controllers (MPCs) providing a simple and elegant solution to string stable ACCs without prediction needs, and v) measurement of the acceleration/deceleration constraints from commercial ACC vehicles, which may restrict string stability. Most of the findings in this dissertation are verified using commercially-available ACC vehicles. The proposed designs are also ready for practical implementation.Ph.D