Source-linked AI summary
Shared Autonomous Vehicle Simulation and Service Design
Reza Vosooghi, Jakob Puchinger, Marija Jankovic, Anthony Vouillon
TL;DR
The paper examines how SAV demand and performance vary with operational design and traveler preferences. It simulates alternative fleets and service strategies in Rouen Normandie, finding that fleet-size choices differ across service configurations and that rebalancing changes several performance measures.
Problem
SAV service design must account for operational configuration and traveler preferences because these factors shape demand and service performance.
Source: §1 · p. 1Method
The study simulates dynamic multimodal SAV demand with user taste variation and compares fleet sizes, capacities, ridesharing, and rebalancing in Rouen Normandie.
Source: §1 · p. 2, §5 · p. 17Results
The best-performing fleet size differs between individual rides and ridesharing, while rebalancing shortens average waiting times by 25-35% for optimum ridesharing fleets.
Source: §5 · p. 18Takeaways & Limitations
Optimum scenarios average 546 to 647 km driven without rebalancing and 707 to 746 km with it, while maximum distance exceeds 900 km, indicating a need for recharging infrastructure.
Source: §5 · p. 18The study applies the same pricing scheme across rides and identifies dynamic pricing as a subject for future work.
Source: §5 · p. 18
Abstract
from arXiv · showhide
Today, driverless cars, as a new technology that allows a more accessible, dynamic and intelligent form of Shared Mobility, are expected to revolutionize urban transportation. One of the conceivable mobility services based on driverless cars is shared autonomous vehicles (SAVs). This service could merge cabs, carsharing, and ridesharing systems into a singular transportation mode. However, the success and competitiveness of future SAV services depend on their operational models, which are linked intrinsically to the service configuration and fleet specification. In addition, any change in operational models will result in a different demand. Using a comprehensive framework of SAV simulation in a multi-modal dynamic demand system with integrated SAV user taste variation, this study evaluates the performance of various SAV fleets and vehicle capacities serving travelers across the Rouen Normandie metropolitan area in France. Also, the impact of ridesharing and rebalancing strategies on service performance is investigated.Research results suggest that the performance of SAV is strongly correlated with the fleet size and the strategy of individual or shared rides. Further analysis indicates that for the pricing scheme proposed in this study (i.e., 20% lower for ridesharing scenario), the standard 4-seats car with shared ride remains the best option among all scenarios. The results also underline that enabling vehicle-rebalancing strategies may have an important effect on both user and service-related metrics. The estimated SAV average and maximum driven distance prove the importance of vehicle range and charging station deployment.
1. Introduction
The paper addresses uncertainty about which SAV service designs will perform effectively, noting that both operational configuration and variation in traveler preferences shape demand. It proposes comprehensive, dynamic multimodal simulations to assess these factors in Rouen Normandie.
1. Introduction
- The study addresses uncertainty about which SAV service models will prevail in efficiency and effectiveness as operators consider new sharing business models.Driverless cars are expected to be shared and accessible, but their operational models remain uncertain.Source: §1 · p. 1
- Fleet size, allocation, relocation, infrastructure, travel time, cost, waiting, detours, and traveler attitudes all affect SAV demand and service performance.The absence of a driver may also concern travelers and lower demand.Source: §1 · p. 1
- The study addresses a gap by simulating dynamic multimodal demand with user taste variation and evaluating operational components and vehicle capacities in Rouen Normandie.It uses MATSim and real-world transport-system data for the case study.Source: §1 · p. 2
2. Prior research
Prior SAV research spans survey-based demand estimates and agent-based simulations, but studies vary in demand dynamics, behavioral representation, and operational scope. The paper identifies vehicle capacity, service costs, rebalancing, and integrated real-world evaluation as underexamined elements.
2. Prior research
- SAV demand research uses surveys for rough stated-preference estimates and agent-based simulation for supply-side studies, including discrete-choice and co-evolutionary approaches.The paper focuses its review on multimodal simulations with dynamic demand responsive to network and traffic conditions.Source: §2 · p. 2
- Earlier studies examined varied replacement scenarios, from Singapore and Stuttgart to Lisbon, using different service assumptions and optimization approaches.Reported outcomes include higher demand with rebalancing, fleet-size-related waiting time, reduced vehicle mileage, and expected congestion and emissions declines.Source: §2 · p. 2
- Prior work includes dynamic pricing and last-mile service studies, but some omit trips under one mile or non-motorized modes, while others omit user-specific attributes.One study found SAV modal share sensitive to value of travel time and fares; another identified trade-offs among fleet size, occupancy, and traveler experience.Source: §2 · p. 3
- Earlier studies often used static or predefined demand, single-mode simulations, or optimization that omitted multimodal mode choice or congestion-dependent travel times.The reviewed gaps motivate a more integrated simulation scope.Source: §2 · pp. 3–4
- Vehicle capacity and its performance effects, service cost, and rebalancing received limited attention; the authors identify these as gaps in comprehensive real-world SAV simulation.Prior comprehensive simulations generally used homogeneous sociodemographic behavior, with one cited exception limited to individual rides (Vosooghi et al., 2019).Source: §2 · p. 4
3. Model specification and set up
The model combines MATSim’s iterative agent-based simulation with dynamic vehicle dispatch and rebalancing, a synthetic Rouen population, varied SAV preferences, and scenario-specific assumptions. It evaluates service and system performance using multiple traveler, fleet, and transport indicators.
3.1. Simulation framework
- MATSim agents execute daily activity plans in dynamic traffic, score outcomes, and re-plan iteratively until plan selection and system scores converge.The framework is extended with DVRP-DRT for dynamic SAV operations.Source: §3.1 · p. 4
3.2. Ridesharing and rebalancing
- A centralized ridesharing optimizer inserts feasible requests to minimize added vehicle work, while periodic rebalancing relocates idle vehicles using expected demand and congestion.The selected dispatch and relocation algorithms may strongly affect performance indicators; more sophisticated heuristics could require extra computation for equilibrium.Source: §3.2 · pp. 4–5
3.3. Inputs and model setup
- The Rouen synthetic population uses census and survey data, fitness-based multilevel controls, and activity chains assigned by socio-professional attributes.The framework figure summarizes synthetic population generation and activity chaining.Source: §3.3 · p. 5, Figure 1 · p. 5
3.4. User taste variation
- SAV utility functions vary across user groups using trust and willingness-to-use factors associated with age, gender, income, and socio-professional attributes.The local survey found men and younger people generally more likely to use SAVs.Source: §3.4 · p. 6
3.5. Simulated scenarios
- Rouen Normandie provides local survey data and a roughly 500,000-person study area simulated at a 10% downscaling rate.The authors note that more extensive downscaling may affect performance evaluation through spatiotemporal supply-demand interactions.Source: §3.5 · p. 6
3.6. Performance metrics
- Fleet sizing balances in-service rate against extra vehicle-kilometers, while detours, waiting, passenger-kilometers, and occupancy characterize user and service outcomes.The study also compares mode shares and car and SAV distances, but does not evaluate environmental impacts or uncertain service and infrastructure costs.Source: §3.6 · pp. 6–7
4. Case study results
Across the case-study scenarios, SAV performance varies with fleet size, ride-sharing, and vehicle capacity. Results show trade-offs among passenger service, fleet use, empty travel, and operating costs.
4.1. Overview
- SAV modal-share gains come from both public transport and car modes, and the fleet-size response differs across service scenarios.The study reports that with large fleets, public transport share decreases significantly relative to car share.Source: Table 2 · p. 7, §4.1 · pp. 7–8
4.2. Fleet size
- 3.5 k SAVs for individual rides, 2.5 k for 2-seats ridesharing, and about 3.0 k for other scenarios are identified as best-performing fleet sizes.Fleet usage can drop by up to 16% as fleet size increases, while the empty-distance ratio changes by at most 3%; un-rebalanced ridesharing fleets do not reach full in-service use.Source: §4.2 · pp. 10–11, Figure 3 · p. 10, Figure 4 · p. 11
- As fleet size grows, the share of employed SAV users falls, while retired people and homemakers use the service more, especially for individual rides.The authors note that user profiles can affect estimates of hourly fleet usage and fleet sizing.Source: §4.2 · p. 11, Figure 5 · p. 12
4.3. The sharing strategy of ride
- Individual rides have the lowest overall PKT across fleet sizes, while ridesharing scenarios have nearly equal PKT for medium fleets of 3.5 k to 4.5 k vehicles.The relative PKT difference between individual and shared rides narrows as fleet size increases.Source: Figure 6 · p. 12, §4.3 · p. 12
4.4. Vehicle capacity
- The standard 4-seats car appears more compatible with ridesharing than the 6-seats minivan, whose sixth seat is almost never used.The extra capacity may still help larger groups and families, and can affect service level.Source: §4.4 · pp. 12–13, Figure 7 · p. 13
4.6. Ridesharing service cost
- For fleets of 2.5 k to 3.5 k SAVs, the initial ridesharing price remains more advantageous than prices 30–40% lower, despite 4–10% higher PKT at lower prices.The PKT gain does not offset the loss of direct income, and EVK can also increase.Source: §4.6 · p. 15, Figure 9 · p. 16
4.7. Rebalancing strategy
- Rebalancing increases modal share, fleet usage, and in-vehicle PKT across scenarios, but also raises empty-distance ratios and driven distances.Average waiting time meaningfully decreases for ridesharing scenarios, while detour and in-vehicle times remain nearly unchanged.Source: §4.7 · p. 16, Table 6 · p. 17
5. Discussion and conclusion
Across the Rouen simulations, SAV performance depends on fleet size and ride strategy; the 3,000-vehicle shared-ride scenario performs best among those considered. Rebalancing changes wait times and travel distances, while the simulated distances underscore the need to consider vehicle range and charging infrastructure.
5. Discussion and conclusion
- SAV modal shares range from 3.1% to 7.6% across fleets of 2,000 to 6,000 vehicles, with individual-ride performance especially sensitive to fleet size.Below the 2,500-vehicle critical size, fleet usage falls 34%; the decrease from 3,000 to 2,500 vehicles is 8%.Source: §5 · p. 17
- Fleet-size effects on waiting time differ by ride strategy: in medium and large ridesharing fleets, average waits change insignificantly, while they decrease for large individual-ride fleets.The ridesharing fleet is not saturated in these cases, whereas individual rides allocate the nearest empty vehicle; in-vehicle and detour-time changes are under four minutes.Source: §5 · p. 17
- Among the considered scenarios, 3,000 vehicles with shared rides is the best option, with relatively high in-vehicle PKT and a low empty-distance ratio.The proposed ridesharing price is 20% below the individual-ride price; reducing it to 40% below does not proportionately improve operational indicators and is not beneficial for the operator.Source: §5 · p. 18
- For optimum ridesharing fleets, rebalancing shortens average waits by 25–35%, while empty-distance ratios rise 42–85% versus 7–17% for in-vehicle PKT.Individual-ride waiting times remain unchanged; detour times are almost unchanged across scenarios.Source: §5 · p. 18
- Without rebalancing, average driven distances for optimum scenarios are 546–647 km; with rebalancing, they increase to 707–746 km, and maximum distance exceeds 900 km.The authors conclude that SAVs will require recharging infrastructure, given the relatively lower range of today’s electric vehicles.Source: §5 · p. 18
- The simulated morning peak shifts from 7–9 a.m. for small fleets to 8–10 a.m. for large fleets, reflecting user trip-pattern and taste variation.The authors identify user differentiation as important to SAV demand modeling and simulation.Source: §5 · p. 18
- The authors identify dynamic pricing, improved rebalancing, and integrating electric SAVs with charging-station locations and vehicle ranges as future work.These are stated opportunities to improve the study’s simulation and service analysis.Source: §5 · p. 18