Shared Autonomous Vehicle (SAV) fleet operations across the Minneapolis-Saint Paul region, with emphasis on empty travel, response times, and no-idling laws over space and time of day

Abstract

Many well-known enterprises are road-testing fully-automated vehicles (AVs), including General Motors, Waymo, Uber, Tesla, and Apple. Most AVs are expected to be used in shared AV (SAV) fleets initially, for daily trip-by-trip use, as an autonomous ride-hailing service. SAVs will allow savings on vehicle ownership and maintenance costs, parking search time, and parking access times. This study micro-simulates passenger travel throughout the Minneapolis-Saint Paul (MSP) region of Minnesota, when relying on a system of SAVs. The extended region includes 9.5 million person-trips per weekday, 7 counties, 2485 traffic analysis zones (TAZs), and about 42,000 roadway links (obtained using OpenStreetMap). An agent-based toolkit, MATSim, allows tracking of individual travelers throughout the day and across their activity locations. The region's metropolitan planning organization, Metropolitan Council, provided all travelers' itineraries, trip purposes, origins, and destinations, along with land use data (jobs and population counts) by TAZ. To simulate SAV assignments to each traveler requesting a trip, along with traveler wait times and arrival times at their destinations, the code form Hörl (2017) - who extended Bischoff and Maciejewski's (2016) MATSim codes - and MATSim's autonomous mobility-on-demand (AMoD) simulator were used here. The SAV fleet size and starting locations were specified before a typical weekday's simulation for 2015. All travel demands sampled from the MSP population must be met by the SAV fleet if they can be met within a pre-specified max-wait-time duration of 1 hour. Travelers are assumed to cancel their SAV request after waiting more than 1 hour. Finally, special SAV parking lots or waiting areas were created to avoid SAVs idling on busy streets in the downtown and other popular locations, between serving trips, to see how such curb-use policies affect wait times and other fleet performance metrics. Using supercomputers, this work simulated 180,000 person-trips and 450,000 person-trips (2% and 5% of the region's 9.2 million daily person-trips) and 480,000 person-trips for the Twin Cities over a 24-hour weekday. Results suggest that the average SAV in this region can serve at most 30 person-trips per day with less than 5 minutes of average wait time for travelers, thus replacing about 10 household vehicles (assuming no one needs to leave the region) but generating another 13 % vehicle-miles traveled (VMT) each day, thereby adding some congestion to the network. By enabling and encouraging active use of for dynamic ride-sharing (DRS), where strangers share rides together, the SAV fleetwide VMT fell, on average, by 17% - and empty VMT (eVMT) fell by 26%, as compared to scenarios without DRS. Interestingly, the 81% and 84% of TAZs with less than 6 minutes average wait times (in the AM and PM peak periods, respectively) are uniformly distributed over this large, 7-county region, suggesting that MSP residents will enjoy similar SAV service levels everywhere (though response times do rise during peak times of day). For the Twin Cities region, most eVMT emerges in the northern and southern subregions, rather than in the cities' CBDs. eVMT and wait times are relatively high during the AM and PM peak periods (6 am to 9 am and 3 pm to 6 pm) but fall significantly during the PM peak period if DRS is offered and actively used by travelers. When compared to idling-at-curb scenarios, the no-idling-on-busy-downtown-road-segment scenarios (using central SAV parking lots) generated 8% more VMT, while eVMT rose by 9 percentage points on average, across all 4 companion scenarios. This study also estimated various energy and emissions savings of SAVs versus the U.S. status quo. Compared to the average household passenger car (a 4-door sedan), which uses 31 miles per gallon, a fleet of 52 mi/gallon hybrid electric SAVs are estimated here to lower the energy demands by 21% and emission related health costs by roughly 30%, sulfur dioxide (SO₂) by 20%, carbon monoxide (CO) by 46%, oxides of nitrogen (NOx) by 30%, volatile organic compounds (VOC) by 48%, particulate matter that is 10 micrometers or less in effective diameter (PM10) by 20%, carbon dioxide (CO₂) by 20% and methane (CH₄) by 35%. Such fleet shifts would save roughly 30% in emissions-related health costs and 64% in energy use.