In this presentation, we show examples of applying agent-based modeling, simulation, and control in relation to modeling the behavioral and learning aspects of drivers and control systems. We start by describing a general modeling framework for estimation and control of adaptive systems. We introduce the general concepts of agent-based modeling and apply the aforementioned framework to train individual agents using reinforcement learning techniques in order to learn the optimal actions in a system. We provide examples of driver modeling as well as adaptive controllers using Q-learning and Neuro-fuzzy systems. We then show an example of emerging system behavior of a multi-agent reservation system to resolve conflicts between vehicular traffic at intersections. In this framework, vehicles reserve space tiles and time slots at an intersection in a first-come-first-served discipline. However, higher priority vehicles are allowed to revoke existing reservations of low priority vehicles, forcing them to re-reserve later time slots. A dynamic algorithm changes the priority of vehicles at appropriate times to account for safety aspects at the junction. Intelligent vehicle agents utilize the information provided by a controller agent and change their trajectories utilizing connected and automated vehicle technologies.
Back to Workshop III: Traffic Control