Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

HRSim: An agent-based simulation platform for high-capacity ride-sharing services

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read HRSim is an open-source, agent-based platform that simulates high-capacity ride-sharing at city scale by jointly modeling pricing, routing, matching, and repositioning.

desk verdict HRSim is a real open-source ride-sharing simulator with a solid modular architecture, but the paper overclaims predictive power and never quantifies 'large-scale'. read the letter →

arxiv 2505.17758 v1 pith:U4INMZK3 submitted 2025-05-23 eess.SY cs.SY

Wang Chen , Hongzheng Shi , Jintao Ke This is my paper · ORCID
classification eess.SYcs.SY
keywords agent-basedsimulationhigh-capacityride-sharingdynamicride-sourcingoperationsintegerlinearprogrammingvehicleroutingrepositioningopen-sourceplatform
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

HRSim is an agent-based simulation platform built to test high-capacity ride-sharing operations on real city road networks and demand data. The paper argues that a single modular, open-source system can jointly simulate pricing, routing, matching, and idle-vehicle repositioning while supporting both pooled and solo trips, and can do so at city scale. If that claim holds, HRSim gives transportation network companies and policymakers a reusable testbed for evaluating operational strategies such as pricing rules, fleet sizing, and matching algorithms before field deployment. The paper demonstrates this utility through applications that quantify carbon emissions, derive scaling laws for service performance, and analyze trade-offs between social welfare and revenue.

What carries the argument

The load-bearing mechanism is the separation of routing from matching inside a modular agent loop. In each step, candidate passengers are grouped by a greedy nearest-neighbor heuristic that builds a pickup-and-delivery route and drops any group whose detour would exceed the promised maximum; only feasible groups become trips. Matching is then an integer linear program maximizing $\sum_{i\in I}\sum_{j\in J} x_{ij} u_{ij}$ with $u_{ij} = \sum_{r\in j} p_r - c_{ij}$, subject to one trip per vehicle and one vehicle per request, solved by commercial solvers. Idle vehicles are repositioned by a second ILP that minimizes total dispatch distance. The decoupling keeps the combinatorial pooling problem from exploding as passenger count grows, and the open-source modular design lets users swap in their own pricing, matching, or repositioning strategies.

What would settle it

Run HRSim on a city where a transportation network company has recorded actual ride-sharing acceptance decisions under known discounts and detour guarantees, then compare the model's predicted share of accepting passengers and resulting service rate with the observed values; a systematic gap would show that the survey-calibrated elasticity module does not transfer.

Watch

Extended reading notes

Core claim

The central claim is that a modular agent-based architecture with decoupled routing and matching makes large-scale, high-capacity ride-sharing simulation practical. HRSim connects real road-network and trip-demand data to a pipeline of six modules: an infrastructure layer, a survey-calibrated pricing module, a routing module that uses Dijkstra shortest paths and a greedy nearest-neighbor heuristic to check whether passengers can share a vehicle, an integer linear programming matcher that assigns vehicles to trips by maximizing total utility, a repositioning module that dispatches idle vehicles by minimizing total travel distance, and a visualization module. Because routing feasibility is resolved before matching, the assignment ILP only sees trips with explicit routes, which keeps each matching step tractable. The paper reports that this design supports city-scale simulations and has already been used to quantify carbon reductions from larger vehicle capacities, to fit scaling laws that predict service rate from system load, and to analyze the revenue and welfare effects of adding ride-sharing to a solo-hailing market.

Load-bearing premise

The pricing module decides whether a passenger joins a shared ride using willingness curves taken from a survey of about 400 people, and HRSim applies those same curves to every simulated passenger in the city; if the surveyed riders are not representative of the city's actual riders, every downstream statistic such as participation rate, service rate, revenue, and emissions will be off.

Editorial extensions

If this is right

  • Researchers can benchmark new ride-sharing algorithms, including reinforcement-learning agents, in a realistic city-scale environment before any field test.
  • The published applications imply that higher-capacity vehicles reduce carbon emissions per passenger-kilometer, with rough reductions of 30 percent for two-passenger, 45 percent for four-passenger, and 50 percent for six-passenger pools relative to solo hailing.
  • The scaling laws fitted from HRSim simulations state that average service rate and average occupancy can be predicted from a single system-load variable, allowing operators to estimate performance in a new city without running a full simulation.
  • The strategy analysis shows that adding ride-sharing to solo hailing can raise service rate, occupancy, and wait-time performance while cutting emissions, but can also reduce revenue if pricing is not designed carefully.
  • Because HRSim is open source, users can extend it to new service modes, road networks, and demand patterns rather than being locked into a closed simulator.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The survey-based price and detour elasticities are a single calibration point; a natural next step would be to re-fit the pricing module on revealed-preference data from actual trip records and test whether simulated participation rates still hold.
  • If the scaling laws generalize beyond the ten cities tested, they could turn ride-sharing performance prediction into a back-of-the-envelope calculation, though the paper does not yet establish that generality.
  • The platform's assumption that vehicles fully comply with assigned routes fits autonomous fleets well, but for human drivers it may overstate matching efficiency; introducing driver acceptance or repositioning reluctance would be a useful stress test.
  • The modular ILP matching formulation is generic enough that the same code could be adapted to other shared-mobility settings, such as shared freight or microtransit, by changing the utility definition.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. HRSim is an open-source, agent-based simulation platform for high-capacity ride-sharing services. The paper describes a six-module architecture (infrastructure, pricing, routing, matching, repositioning, visualization) that integrates real-world road networks and trip data, supports both solo-hailing and ride-sharing, and uses ILP-based matching/repositioning with heuristic routing. The central claims are that this modular design enables large-scale, city-scale simulations and provides a reusable testbed for pricing, matching, repositioning, carbon-emission quantification, and strategy evaluation. Quantitative demonstrations are not carried out directly in the paper; instead, results from three companion papers [18]–[20] are referenced to illustrate the platform's utility.

Significance. If the platform performs as claimed, it would make a valuable contribution: an open-source, modular, city-scale testbed for ride-sharing research is genuinely useful, particularly because HRSim decouples routing from matching, uses standard ILP formulations, and supports multiple service modes. The open-source code availability and the companion applications to carbon emissions, scaling laws, and welfare analysis are concrete strengths. However, the present manuscript does not itself supply the evidence needed to substantiate the central scalability and realism claims: no runtime or fleet-size numbers, no validation of the survey-calibrated passenger behavior model, and no direct comparison with existing platforms are reported. The value is therefore conditional on additional validation and clearer quantitative reporting.

major comments (3)
  1. [Sec. II-B, Fig. 2] The passenger participation model is calibrated from approximately 400 questionnaires and then applied unchanged to every simulated passenger in any city, including cities with very different rider populations (e.g., NYC, Chicago) that provide the demand data. The paper provides no evidence that this small sample is representative of those cities, no comparison against observed shared-ride acceptance behavior in the corresponding TNC datasets, and no sensitivity analysis around the elicited price and detour elasticities. Since participation rate is an input to routing, matching, and repositioning, a systematic bias here propagates into every downstream metric, including service rate, occupancy, revenue, waiting time, and emissions. The testbed claim therefore depends on an unvalidated behavioral assumption that should be addressed, either by validating the elasticities against real shared-trip data or by presenting sensitivity analyses over the elasticity parameters.
  2. [Sec. III-B, Eq. (12)] The scaling law in Eq. (12) is presented as predicting the performance of ride-sharing in other cities, but α and β are fitted parameters calibrated to the ten-city HRSim results, and the parameter list also contains an apparent typo: "α = 1.8 and β = 1.7 when C = 2, α = 2.7 and β = 3.5 when C = 2, and α = 2.4 and β = 3.6 when C = 6". The second case should likely be C = 4, not C = 2. As written, the match to the ten cities is a fitted curve, not an independent prediction, and no goodness-of-fit or cross-validation statistics are reported. The authors should correct the typo, explicitly state that Eq. (12) is an empirical fit, and, if they intend to claim predictive power for unseen cities, provide holdout validation.
  3. [Sec. I and Sec. III] The paper's central claim that HRSim "can achieve large-scale simulations" and "perform simulations at a city scale" is never quantified in this manuscript. There are no runtime measurements, no reported fleet sizes, city sizes, request rates, or simulation durations, and no comparison with the existing platforms listed in Table I. Without such numbers, the scalability claim that distinguishes HRSim from prior work cannot be evaluated. The authors should report concrete experimental settings (e.g., number of vehicles, demand rate, road-network size, wall-clock time per simulated hour) and ideally include a scaling experiment that shows performance as system size grows.
minor comments (4)
  1. [Sec. II-B] The sentence "we primally focus on the discount and induced detour of ride-sharing" contains a typo: "primally" should be "primarily".
  2. [Sec. II-E, Eq. (6)] The definition of d_{vw} reads "the distance from the location of vehicle v to where passenger w is"; this should be "to the location of passenger w" for clarity and consistency with the subsequent constraint description.
  3. [References] Reference [14] misspells the author name as "Dijksta"; it should be "Dijkstra".
  4. [Table I] The checkmark criteria in Table I are not defined (e.g., what qualifies as "Large-scale" or "Agent-based"), and some entries appear surprising (e.g., UXsim marked as agent-based); please clarify the criteria or cite the specific features that justify each checkmark.

Circularity Check

1 steps flagged · score 6.0 of 10

The scaling-law 'prediction' in Sec. III-B is a fitted curve evaluated at u=3, so the ten-city match is goodness-of-fit rather than an independent forecast; this is the one concrete circular step.

  1. fitted input called prediction [Section III-B 'Ride-sharing scaling', Eqs. (10)-(12) and the paragraph beginning 'These scaling laws can accurately reproduce...']
    "These scaling laws can accurately reproduce the experimental results across ten cities, indicating that they could be used to predict the performance of ride-sharing services in other cities [19]. ... For instance, when the system load is high (e.g., u = 3), high-capacity ride-sharing services with a vehicle capacity of 4 passengers can accommodate 72% of the demand on average."

    Eq. (12) contains fitted parameters alpha and beta, calibrated in [19] to the same ten-city HRSim simulation outputs that the present paper says it 'accurately reproduce[s]'. Reproducing the data used to estimate a regression is fitting, not prediction. The illustrative 72% and 50% service-rate numbers are obtained by evaluating this fitted curve at u = 3 for the corresponding vehicle capacities, so those 'predicted' values are forced by the fitted equation by construction. No out-of-sample city is tested, so the claim that the law 'could be used to predict performance in other cities' rests entirely on the in-sample fit rather than on any independent verification.

full rationale

The paper's central contribution is the open-source HRSim platform itself, whose existence and architecture are independently checkable from the released code and from the modular pipeline described in Sec. II. That core claim is not circular: road networks, demand data, pricing survey, Dijkstra routing, ILP matching, and repositioning are all specified inputs and standard algorithmic components, not quantities derived from the simulation outputs. The pricing-survey limitation (roughly 400 questionnaires) and the reliance on the authors' own prior applications [18]–[20] are weaknesses or self-citations, but they do not reduce the platform's construction to its conclusions. The one concrete circular step is in the 'Ride-sharing scaling' application: the scaling law of Eq. (12) is presented as predictive, yet its alpha and beta are fitted to the very ten-city results it claims to reproduce, and the illustrative projection at u=3 is simply an evaluation of that fitted curve. This is a fitted-input-called-prediction pattern, producing a partial circularity score of 6 rather than a fully circular derivation. The remaining self-citations are not load-bearing enough to raise the score further because the platform is open source and can be tested independently.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The platform's modeling choices introduce several dependent quantities. The scaling-law parameters alpha and beta are explicitly fitted to HRSim simulation output in the cited prior work and are then used to state predictive scaling behavior. The pricing module relies on a survey-based elasticity calibration that is not published as data. The simulation assumes full vehicle compliance and deterministic route execution.

free parameters (2)
  • Scaling-law parameters alpha(C) and beta(C) = alpha=1.8, beta=1.7 for C=2; alpha=2.7, beta=3.5 for C=4 (paper prints C=2 again); alpha=2.4, beta=3.6 for C=6
    Appear in Eq. (12) and are fitted to HRSim simulation output across ten cities in the authors' prior paper [19]; they are used to state the service-rate scaling law.
  • Passenger price and detour elasticity curves = not reported numerically
    Calibrated from about 400 questionnaires in Sec II-B (Fig. 2) and used in the pricing module to decide whether a simulated passenger accepts ride-sharing. The questionnaire data are not provided.
assumptions (4)
  • domain assumption Vehicles are fully compliant and always follow the routes planned by the platform
    Stated in Sec II-A. Justified by appeal to autonomous driving, which limits realism for human-driven ride-hailing fleets.
  • domain assumption The nearest-neighbor routing heuristic is sufficiently accurate, over 95% agreement with enumeration, to determine which passengers can be pooled
    Sec II-C cites [18] for this accuracy claim but does not re-verify it here; the routing feasibility check underlies the matching module.
  • domain assumption Demand data from public trip datasets such as NYC and Chicago are representative enough to simulate ride-sharing operations in a city
    Sec II-A assumes TNC or open-source trip records can represent demand; no comparison to actual ride-sharing trip data is given.
  • standard math The ILP matching and repositioning models capture the relevant assignment constraints
    Eqs. (2)-(5) and (6)-(9) are standard assignment formulations solved with CPLEX or Gurobi; no proof of completeness is given, but the constraint structure is conventional.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HRSim: An agent-based simulation platform for high-capacity ride-sharing services." pith.science (2026). https://pith.science/paper/U4INMZK3

@misc{pith2026250517758,
  author       = {Pith},
  title        = {Pith review of: HRSim: An agent-based simulation platform for high-capacity ride-sharing services},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U4INMZK3}},
  note         = {Machine review of arXiv:2505.17758}
}
read the original abstract

The rapid growth of ride-sharing services presents a promising solution to urban transportation challenges, such as congestion and carbon emissions. However, developing efficient operational strategies, such as pricing, matching, and fleet management, requires robust simulation tools that can replicate real-world dynamics at scale. Existing platforms often lack the capacity, flexibility, or open-source accessibility needed to support large-scale, high-capacity ride-sharing services. To address these gaps, we introduce HRSim, an open-source, agent-based High-capacity Ride-sharing Simulator. HRSim integrates real-world road networks and demand data to simulate dynamic ride-sharing operations, including pricing, routing, matching, and repositioning. Its module design supports both ride-sharing and solo-hailing service modes. Also, it includes a visualization module for real-time performance analysis. In addition, HRSim incorporates integer linear programming and heuristic algorithms, which can achieve large-scale simulations of high-capacity ride-sharing services. Applications demonstrate HRSim's utility in various perspectives, including quantifying carbon emissions, scaling ride-sharing performance, evaluating new strategies, etc. By bridging the gap between theoretical research and practical implementation, HRSim serves as a versatile testbed for policymakers and transportation network companies to optimize ride-sharing systems for efficiency and sustainability.

Figures

Figures reproduced from arXiv: 2505.17758 by the authors.

Figure 1
Figure 1. Architecture of HRSim including (a) Infrastructure, (b) pricing module, (c) routing module, (d) matching module, (e)repositioning module, and (f) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Passengers’ Price and detour elasticity calibrated with survey data. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Quantified carbon emissions of non-ride-sharing (NS) and ride-sharing [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Implications of incorporating ride-sharing services. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. RideGym: A Standardized Interface for Real-World Large-Scale Ride-Sharing System

    cs.MA 2026-07 accept novelty 5.5 of 10

    RideGym provides the first open, algorithm-agnostic Gym interface for large-scale ride-sharing order dispatch and shows exploration noise can reverse MARL performance rankings.

Reference graph

Works this paper leans on

20 extracted references · 18 canonical work pages · cited by 1 Pith paper

  1. [18]

    Quantifying traffic emission reduc- tions and traffic congestion alleviation from high-capacity ride-sharing,

    W. Chen, J. Ke, and X. Chen, “Quantifying traffic emission reduc- tions and traffic congestion alleviation from high-capacity ride-sharing,” Transportmetrica B, Transp. Dyn. , vol. 12, no. 1, p. 2423235, 2024

    work page 2024

  2. [20]

    Development dilemma of ride-sharing: Revenue or social welfare?

    W. Chen, G. Huang, and J. Ke, “Development dilemma of ride-sharing: Revenue or social welfare?” arXiv:2412.08801, 2024

    work page Pith review arXiv 2024

  3. [1]

    On-demand high-capacity ride-sharing via dynamic trip-vehicle assignment,

    J. Alonso-Mora, S. Samaranayake, A. Wallar, E. Frazzoli, and D. Rus, “On-demand high-capacity ride-sharing via dynamic trip-vehicle assignment,” Proc. Natl. Acad. Sci. USA , vol. 114, no. 3, pp. 462–467, 2017

    work page 2017

  4. [2]

    Exact matching of attractive shared rides (ExMAS) for system-wide strategic evaluations,

    R. Kucharski and O. Cats, “Exact matching of attractive shared rides (ExMAS) for system-wide strategic evaluations,” Transp. Res. Part B Methodol., vol. 139, pp. 285–310, 2020

    work page 2020

  5. [3]

    Q4 2023 Prepared Remarks,

    D. Khosrowshahi and P. Mahendra-Rajah, “Q4 2023 Prepared Remarks,”2024. https://s23.q4cdn.com/407969754/files/doc earnings/ 2023/q4/transcript/Uber-Q4-23-Prepared-Remarks.pdf. Access date: April 23rd, 2025

    work page arXiv 2023

  6. [4]

    Coordinating ride-sourcing and public transport services with a reinforcement learning approach,

    S. Feng, P. Duan, J. Ke, and H. Yang, “Coordinating ride-sourcing and public transport services with a reinforcement learning approach,” Transp. Res. Part C Emerg. Technol. , vol. 138, p. 103611, 2022

    work page 2022

  7. [5]

    Efficient large-scale fleet management via multi-agent deep reinforcement learning,

    K. Lin, R. Zhao, Z. Xu, and J. Zhou, “Efficient large-scale fleet management via multi-agent deep reinforcement learning,” in Proc. 24th ACM SIGKDD Int. Conf. Knowledge Discovery & Data Mining , 2018, pp. 1774–1783

    work page 2018

  8. [6]

    A multi- functional simulation platform for on-demand ride service operations,

    S. Feng, T. Chen, Y . Zhang, J. Ke, Z. Zheng, and H. Yang, “A multi- functional simulation platform for on-demand ride service operations,” Commun. Transp. Res. , vol. 4, p. 100141, 2024

    work page 2024

Show all 20 references
  1. [7]

    UXsim: An open source macroscopic and mesoscopic traffic simulator in Python–a technical overview,

    T. Seo, “UXsim: An open source macroscopic and mesoscopic traffic simulator in Python–a technical overview,” arXiv:2309.17114, 2023

    2023 arXiv

  2. [8]

    Multi-agent simulation for planning and designing new shared mobility services,

    G. Inturri, M. Le Pira, N. Giuffrida, M. Ignaccolo, A. Pluchino, A. Rapisarda, and R. D’Angelo, “Multi-agent simulation for planning and designing new shared mobility services,” Res. Transp. Econ. , vol. 73, pp. 34–44, 2019

    2019

  3. [9]

    A simulation framework for real-time assessment of dynamic ride sharing demand responsive transportation models,

    M. P. Linares, L. Montero, J. Barcelo, and C. Carmona, “A simulation framework for real-time assessment of dynamic ride sharing demand responsive transportation models,” in Proc. 2016 Winter Simulation Conf. (WSC), Washington, DC, USA, 2016, pp. 2216–2227

    2016

  4. [10]

    Hybrid operations of human driving vehicles and automated vehicles with data-driven agent- based simulation,

    F. Yao, J. Zhu, J. Yu, C. Chen, and X. (M.) Chen, “Hybrid operations of human driving vehicles and automated vehicles with data-driven agent- based simulation,” Transp. Res. Part D Transp. Environ. , vol. 86, p. 102469, 2020

    2020

  5. [11]

    Simulating on-demand ride services in a Manhattan-like urban network considering traffic dynamics,

    Z.-J. Ding, Z. Dai, X. (M.) Chen, and R. Jiang, “Simulating on-demand ride services in a Manhattan-like urban network considering traffic dynamics,” Physica A Stat. Mech. Appl. , vol. 545, p. 123621, 2020

    2020

  6. [12]

    FLEETPY: A modular open-source simulation tool for mobility on-demand services,

    R. Engelhardt, F. Dandl, A.-A. Syed, Y . Zhang, F. Fehn, F. Wolf, and K. Bogenberger, “FLEETPY: A modular open-source simulation tool for mobility on-demand services,” arXiv:2207.14246, 2022

    2022 arXiv

  7. [13]

    Towards a very large scale traffic simulator for multi-agent reinforcement learning testbeds,

    Z. Hu, C. Zhuge, and W. Ma, “Towards a very large scale traffic simulator for multi-agent reinforcement learning testbeds,” in Proc. 2022 IEEE 25th Int. Conf. Intell. Transp. Syst. (ITSC) , Macau, China, 2022, pp. 363–368

    2022

  8. [14]

    A note on two problems in connexion with graphs,

    E. W. Dijksta, “A note on two problems in connexion with graphs,” Numer. Math, vol. 1, pp. 269–271, 1959

    1959

  9. [15]

    Open Street Map,

    OSM, “Open Street Map,” http://www.openstreetmap.org/. Access date: April 23rd, 2025

    2025

  10. [16]

    TLC Trip Record Data,

    TLC, “TLC Trip Record Data,” https://www.nyc.gov/site/tlc/about/ tlc-trip-record-data.page. Access date: April 23rd, 2025

    2025

  11. [17]

    Transportation Network Providers– Trips (2018–2022),

    Chicago Data Portal, “Transportation Network Providers– Trips (2018–2022),” https://data.cityofchicago.org/Transportation/ Transportation-Network-Providers-Trips-2018-2022-/m6dm-c72p/ about data. Access date: April 23rd, 2025

    2018

  12. [19]

    Scaling laws of dynamic high- capacity ride-sharing,

    W. Chen, L. Yang, X. Chen, and J. Ke, “Scaling laws of dynamic high- capacity ride-sharing,” Transp. Res. C, Emerg. Technol. , vol. 174, p. 105064, 2025

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.