Resumen de Managing algorithmic drivers in a blocked-lane scenario
Due to the emergence of new technologies, algorithm-assisted drivers are close to becoming a reality. In this thesis, different aspects of managing such drivers in a blocked-lane scenario are discussed. The first chapter presents an algorithm for the optimal merging of self-interested drivers. The optimal policy can include undesirable velocity oscillations. We propose measures for a central planner to eradicate them, and we test the efficiency of our algorithm versus popular heuristic policies. In the second chapter, a mechanism for positional bidding of the drivers is developed. It allows trading of highway positions of the drivers with heterogeneous time valuations, resulting in a socially beneficial outcome. The final chapter presents a deep learning policy for centralized clearing of the bottleneck in the shortest time. Its use is fast enough to allow future operational applications, and a training set consists of globally optimal merging policies.
