Planning for a Shared Automated Transportation Future

Abstract

Vehicle automation represents the greatest revolution in transportation since the automobile itself. However, the greater the potential impact of a new technology, the harder the future is to predict; the nature of revolutionary advances is that they are not incremental, and they will be used in far different ways than what came before them. For automated vehicles, this could include wholesale shift towards shared automated vehicles (SAVs), similar to self-driving taxis. While SAVs are not a panacea for problems of the transportation system, this dissertation is based on the premise that an SAV future is far more desirable than one dominated by privately owned and operated AVs (PAVs). Automation in concert with private ownership would encourage more sprawling development, more vehicle kilometers traveled, more congestion, more emissions, and more inequality.

This dissertation seeks to understand and model the conditions under which SAVs are more likely to succeed. It constructs a simplified and flexible agent-based model to test system performance under a wide variety of situations, including varying fleet size, urban density, and urban form. By performing sensitivity analyses on these independent variables, this model is able to identify tipping points and other non-linear performance trends. For example, when fleet sizes increase, overall wait times and relocation percentages (the time/distance the vehicle must travel while empty) decrease, albeit at sharply reducing rates. As long as fleet sizes are sufficiently large to avoid queuing, which occurs when there are more trip requests than available vehicles, further fleet size increases does not substantially improve performance. When urban density increases, overall SAV system performance also increases at decreasing rates. SAV systems do not appear to be viable for densities much lower than approximately 500 people per km2. However, there is not a huge increase in performance when going from medium density (e.g. Ann Arbor) to high density (e.g. Manhattan); SAV systems could work in both kinds of places. Varying urban form has a somewhat more complex relationship with SAV performance. The most important points are that is that trips to the less dense outskirts of a city are more difficult and expensive to serve than those in the city center, and that once trip distance is controlled for, the benefits from serving the denser central areas are more than outweighed by the costs of having to serve the outskirts when compared to a base city with constant density throughout.

The final modelling runs examine mode choice, comparing SAVs with transit and PAVs across different urban densities. These results suggest that SAVs obtain their greatest mode share over medium densities between approximately 500 and 4,000 people per km2, but never more than 13% of all trips. While on their own, SAVs are not sufficient to break the dominance of private vehicles, they can act as a supplement to transit, helping to provide rides that would be otherwise too difficult or expensive for transit to serve. A combination of transit and SAVs can encourage people to go car-free, providing superior service to either of the two alone. To encourage greater SAV usage, the gold standard is public-private partnerships combining transit and SAVs, but other enabling policies include requiring data sharing from ridesourcing companies, regulations favoring sharing over private ownership (e.g. pricing private use and parking), and establishing guiding principles that include resilience, sustainability, equity, and accessibility.