Impact of Autonomous Vehicles and Ride-Sharing on Urban Transportation

This thesis investigates how autonomous vehicles and ride‑sharing influence urban transportation, using the SATURN mesoscopic modelling suite to quantify effects on capacity, delay, routing, and network efficiency. The study updates core SATURN inputs to represent AV behaviour, including saturation flow, gap acceptance, stacking capacity, speed–flow relationships, and generalised cost, and frames the analysis within a static user‑equilibrium assignment to isolate supply‑side effects of automation.
The methodology progresses from controlled junction tests to a toy network and then a city‑scale application for York. Nine scenarios examine unopposed and opposed movements at priority junctions, turning at signals, and network‑wide impacts under different AV penetration shares, with a parallel comparison between a flow‑theory approach and a PCU‑based treatment of AVs. An automated workflow iteratively recalculates node‑level LSAT and GAP and link‑level STACK from the assigned AV share, using SATDB and re‑assignment until stability is reached.
Results show substantial capacity gains. In all‑AV conditions, unopposed turns increase by about 95 percent and signalised turning capacity rises by about 42 percent, which aligns with but slightly exceeds prior benchmarks because the modelling also reduces safety distance. In the toy network, these capacity changes cut total travel time and queues markedly, with large reductions in over‑capacity delay. In the City of York model, increasing AV penetration improves city‑wide performance: total travel time falls by roughly 4 percent at 25 percent AV and about 12 percent at 100 percent AV, average speed rises by about 4 to 12 percent, and average queues fall by about 16 to 50 percent, while total distance changes little. Benefits concentrate on inner‑city signalised corridors where shorter headways reduce lost time, while strategic corridors that run near free flow change little. The flow‑theory outcomes closely match a PCU factor of about 0.7 for AVs across both toy and city networks, supporting use of either approach for planning analysis when calibrated appropriately. Varying AV values of time and operating cost produces only minor network‑wide differences under the tested settings.
Finally, the thesis develops a fast, non‑integer ride‑matching algorithm that aggregates paths by zone sequence and matches rides within a model hour. It integrates with multi‑path assignments, estimates impacts on delay and capacity at city scale, and offers a base for future Shared AV modelling.