In the past few years, there has been much research in the field of Autonomous Vehicles (AV). If AVs are implemented in our daily lives, this could have many advantages. Before this can happen, safe driver models need to be designed which control the AVs. One technique that is suitable to create these models is Reinforcement Learning (RL). A problem here is that an RL agent usually needs to execute random actions during training, which is unsafe when driving an AV. Two shields are proposed to solve this problem: a Safety Checking Shield (SCS) and a Safe Initial Policy Shield (SIPS). The SCS checks whether an action is safe by predicting the future state after taking that action and checking whether that future state is safe. The SIPS checks whether an action is safe by comparing it to a safe action from a Safe Initial Policy. Based on the safety of the current state and this action, a safe range of actions is created in which the chosen action must fall. Furthermore, two shield-based learning techniques are proposed which are part of the RL algorithm and allow the agent to learn to avoid proposing actions that would be overruled by a shield. For the first method, experiences are fabricated, and for the second method, an alternative loss function is adopted. In the CARLA driving simulator, two scenarios were created to test the systems. In the first scenario, the agent needs to learn to drive straight, and in the second scenario, it needs to learn to not hit other vehicles on a straight road. The two shields are built around a Double Deep Q-Network (DDQN) and compared to it. It is shown that both shielding systems have zero collisions during training and execution, while having a similar or even better performance in terms of efficiency in comparison to the baseline DDQN. Furthermore, it is shown that both shield-based learning techniques effectively enable the agent to learn to not propose unsafe actions.

Urban areas are seeing influx of population and therefore are experiencing increasing stress on the systems currently in place. In traffic networks, a larger population means that their are inevitably a short term increase in mobility demands. To accommodate this, governments and and the private sector are proposing new solutions for mobility. One such focus is the introduction of new transport modes. A common classification for these modes is Mobility-as-a-Service (MaaS), which includes modes such as carsharing, shared last-mile transportation like bicycles, and ridesharing. MaaS also overlaps with other new modes such as autonomous and electric vehicles. Modeling these types of modes and their subsequent interactions with each other and traditional modes requires more complex modeling techniques than have been traditionally used in transport modeling.

The purpose of this thesis is to propose a method for using agent-based modeling and simulation (ABMS) to assess traffic network resilience, as ABMS has the dynamic qualities that match well with the time variant nature resilience.. This gap is especially prevalent when considering novel modes of transport, such as ridesharing. Data from the metropolitan region of Rotterdam and The Hague (Metropoolregio Rotterdam Den Haag or MRDH) is used for a case study on the corridor between Rotterdam and The Hague. The goal is to prove feasibility of a method for assessing resilience in traffic networks using ABMS.

A resilience framework for urban mobility is proposed that uses six categories to characterize resilience: (1) reflective, redundant, flexible, resourceful, inclusive, and integrated. Using this framework as a guide, three metrics are proposed for measuring resilience using agent-based simulation: origin-destination (OD) travel time, link travel time, and link volume. These metrics are used to compare scenarios that either include a disturbance in the network or do not, which in this case occurs for a 30 minute period on the A4 roadway between Rotterdam and The Hague. While the OD travel time metric is limited in its usability, the link travel time metric makes apparent the the recovery time to achieve normal operating conditions after a disturbance. In the presented case study, scenarios that included ridesharing had worse recovery time then the car only scenario, as well as a higher maximum travel time across the disturbed link. The link volume metric contextualizes these results, showing that while the overall volume throughout the simulation is lower for the ridesharing cases, the volume during the disturbance across the disrupted portion of the A4 roadway is higher. The higher volume shows why the travel time is higher with the presence of ridesharing in the disturbance scenario. These results are subject to the limitations of the model, though, which include dynamic routing that may not avoid the disturbed portion of the network when the disturbance occurs.