Operations of Shared Autonomous Vehicle Fleet for Austin, Texas, Market

The advent of fully self-driving vehicles promises to revolutionize transportation, leading to the emergence of shared autonomous vehicles (SAVs), essentially driverless taxis. This study investigates the potential impact of SAVs at a low market penetration level in Austin, Texas, a city with a well-defined transportation network and trip data. The research focuses on understanding operational implications, including fleet sizing, wait times, vehicle miles traveled (VMT), and the potential for increased efficiency compared to existing transportation modes such as personal vehicles, taxis, and ride-sharing services. The significance of this research stems from the need to anticipate and prepare for the operational challenges and benefits of widespread SAV adoption. Understanding how SAVs function in a realistic urban environment will guide policy decisions and infrastructure planning for future transportation systems.

The paper reviews existing literature on shared autonomous vehicles, comparing them to car-sharing programs and traditional taxi services. It cites previous research by Kornhauser et al. (on aTaxi systems), Burns et al. (on cost-effectiveness of SAVs), and Fagnant and Kockelman (on SAV cost and return on investment). The authors highlight the advantages of SAVs over existing modes, including unoccupied travel to waiting passengers and proactive relocation based on anticipated demand, which can outweigh increased acquisition and rental costs. The literature review sets the stage for the current study by emphasizing the need for detailed network-based simulations to assess the real-world operational implications of SAVs.

The study utilizes a detailed, time-dependent transportation network of Austin, Texas, derived from the Capital Area Metropolitan Planning Organization (CAMPO). A synthetic population of trips was generated based on CAMPO's regional travel demand model and supplemented with departure time data from Seattle (due to limitations in the Austin data). The simulation uses MATSIM, an agent-based, dynamic traffic simulation software, to obtain link-level hourly average travel times reflecting real-world congestion. A 100,000-trip subset within a central 12-mi x 24-mi geofence was selected for SAV simulation. The simulation itself is a custom C++ program with four key modules: 1. **SAV Location and Trip Assignment:** Uses a modified Dijkstra's algorithm to assign the nearest available SAV to waiting travelers within a 5-minute travel time constraint, with a waitlist and expanding search radius for longer wait times. 2. **SAV Fleet Generation:** Dynamically generates new SAVs if any traveler waits 10 minutes without an assigned vehicle, establishing the initial fleet size for the main simulation. 3. **SAV Movement:** Simulates SAV movement on the network, considering time-dependent travel times and incorporating 1-minute pickup and drop-off times. 4. **SAV Relocation:** Implements a block balancing strategy to reposition SAVs based on the imbalance between available vehicles and anticipated demand in 2-mi x 2-mi blocks. Equation 1 calculates block balance, comparing the share of SAVs in a block to the share of expected demand. This relocation optimizes wait times while accounting for increased VMT due to unoccupied travel. The simulation analyzes various metrics, including wait times, vehicle occupancy, unoccupied VMT, and travel speeds, comparing them to existing transportation modes and assessing the potential environmental impact (emissions and energy consumption).

The simulation results indicate that a fleet of approximately 1977 SAVs is needed to serve the sampled trips within the geofence. Each SAV served an average of 28.5 trips per day. Based on this, each SAV could replace roughly nine conventional vehicles. Average wait times were only one minute, with 94.3% of travelers waiting less than five minutes. However, SAV operations resulted in an 8% increase in VMT due to repositioning for efficiency. Average travel distance-weighted speed was 43.6 mph, while time-weighted average speed (considering total VMT divided by VHT) was 26.1 mph. This reflects the impact of periods with slower speeds during peak hours. A comparison with New York City taxis highlights the potential for significantly improved service and lower unoccupied VMT with SAVs. Simulations also explored electric vehicle implications. The high daily travel distances per SAV (average 174 miles) pose a challenge for current battery electric vehicles (BEVs), requiring considerations for charging infrastructure and potentially favoring plug-in hybrid electric vehicles (PHEVs) or rapid-charging solutions. The analysis also looked at emissions. Comparing the simulated SAV fleet to the existing U.S. light-duty vehicle fleet, results showed significant reductions in greenhouse gas emissions, sulfur dioxide, and particulate matter, even with the increased VMT. The reduction in cold starts also contributed significantly to emission reductions. A comparison between the Austin network-based simulation and previous grid-based simulations showed similar environmental outcomes. While the grid-based model suggested even higher vehicle replacement rates (11.76:1) and lower wait times, it also had a significantly higher unoccupied VMT (32% versus 8%). This difference is attributed to differing assumptions in the models, including trip density, travel speeds, and network structure. The high utilization rate of SAVs (63,335 miles per year) makes them more prone to frequent replacement (every 3-5 years), allowing for regular upgrades to automation technology.

The findings demonstrate the potential for SAVs to significantly improve the efficiency and sustainability of urban transportation. The high vehicle replacement rate shows that SAVs can dramatically reduce the number of privately owned vehicles, leading to reduced parking demand and potential land use improvements. While the increased VMT due to SAV repositioning is a concern, the significant reductions in emissions and cold starts, along with the potential for using smaller, more efficient vehicles, largely offset this negative impact. The comparison between the network-based and grid-based simulations highlights the importance of using realistic network data and accounting for spatial variations in demand when modeling SAV operations. Future research should focus on optimizing relocation strategies, exploring alternative vehicle types (e.g., electric), examining the impact of varying market penetration rates, and integrating SAVs into broader multimodal transportation planning models. The study’s limitations highlight the need for further research to address the complexities of real-world SAV deployment.

This study provides valuable insights into the operational characteristics and potential impact of SAV fleets in a realistic urban setting. The results suggest that SAVs could significantly improve urban transportation by replacing a substantial number of conventional vehicles, leading to reduced congestion, emissions, and parking demand. The increased VMT from unoccupied travel represents a trade-off with improved service, highlighting the need for further research into optimal relocation strategies. Future work should explore the impact of higher market penetration rates, examine the integration of different vehicle types (e.g., electric), and incorporate SAVs into multimodal transportation planning.

The study's limitations include the focus on a single day's simulation within a geofenced area of Austin, potentially underestimating the impact of trips originating or ending outside the geofence. The use of Seattle departure time data might introduce some uncertainty. Furthermore, the model's assumptions about traveler behavior and SAV operational parameters may affect the generalizability of results. The simulation did not fully capture the dynamic interactions between SAVs and other modes of transportation. Further research is needed to address these limitations and refine the understanding of SAV operations in more complex scenarios.