Multi-agent Reinforcement Learning for Motorway Ramp Merging – TAKEN

So-called ‘capacity drop’ is a phenomenon that sometimes occurs when joining vehicles interrupt the flow of traffic on a motorway, effectively reducing the capacity of the road. To address this problem, this project will investigate the design of a cooperative motorway on-ramp merging algorithm for automated vehicles. From the perspective of vehicles joining the motorway, the goal is to learn appropriate trajectories for on-ramp vehicles to join the motorway. From the perspective of vehicles travelling on the motorway, the goal is to learn compatible trajectories for on-road vehicles to create safe gaps into which vehicles that want to join the highway can move. The algorithm should be optimal (i.e., ensure that joining vehicles can always enter the first available space on the motorway that they can reach given their arrival time and speed without disrupting the flow of traffic on the motorway) and safe (i.e. ensure no collisions between vehicles occur). Moreover, the algorithm should accommodate heterogeneous vehicles (i.e., with different acceleration and deceleration profiles and other characteristics) and different ramp and motorway layouts.