pith. sign in

arxiv: 2605.17426 · v1 · pith:EO5UX4E6new · submitted 2026-05-17 · 💻 cs.MA · cs.LG

Human-Flow Digital Twin for Predicting the Effects of Mobility Introduction on Visitor Circulation

Chiharu Shima , Haruki Yonekura , Fukuharu Tanaka , Tatsuya Amano , Hirozumi Yamaguchi This is my paper

Pith reviewed 2026-05-19 22:34 UTC · model grok-4.3

classification 💻 cs.MA cs.LG
keywords human flowdigital twinmulti-agent simulatormobility introductionvisitor circulationmulti-layer perceptrondestination choicespatial population distribution
0
0 comments X

The pith

A digital twin of visitor flows predicts mobility introduction effects by adjusting distances and attractiveness in a trained multi-agent simulator.

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

The paper establishes a framework for a human-flow digital twin that trains agents on pre-intervention data to model destination choices based on location, distances, and spot attractiveness. Once trained, the simulator represents mobility measures as parameter changes and generates predictions of altered visitor counts and circulation patterns. In evaluation at Wakayama Castle Park, a multi-layer perceptron version reproduced observed post-introduction spatial distributions with cosine similarity above 0.7. A sympathetic reader would care because this suggests a way to forecast intervention impacts without collecting complete new flow datasets after each change.

Core claim

The framework extracts pre-intervention human-flow data along with inter-spot distances, spot attractiveness, and travel volumes to train each agent's decision model as a multi-layer perceptron. Mobility introduction measures are then expressed as modifications to inter-point distances or spot attractiveness inside the simulator, allowing reproduction of the resulting human flows and quantification of effects such as shifts in visitor numbers and circulation. On measured data with and without mobility introduction at Wakayama Castle Park, the approach yielded cosine similarity exceeding 0.7 for spatial population distributions.

What carries the argument

Multi-agent simulator whose agents employ a multi-layer perceptron decision model trained to map current location and environmental factors to next destination choice.

If this is right

  • Changes in visitor counts at individual spots after mobility introduction can be quantified by running the adjusted simulator.
  • Circulation patterns and overall flow shifts can be predicted from pre-intervention measurements alone.
  • The method replicates real flow changes observed in a park setting with cosine similarity exceeding 0.7 when using the multi-layer perceptron model.
  • Effects of different mobility measures can be compared by varying the distance or attractiveness parameters in the same trained model.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach could lower the cost of evaluating mobility interventions by reducing reliance on repeated full-scale field measurements after each change.
  • If the decision model proves robust across sites, planners might test mobility options in locations without prior full datasets.
  • Incorporating additional inputs such as time-varying factors into the agent model might extend the framework to more dynamic environments.

Load-bearing premise

That modifying only inter-spot distances or spot attractiveness inside the trained simulator is sufficient to model the behavioral effects of real mobility introduction measures, without other unmodeled factors altering visitor choices.

What would settle it

Measure actual post-mobility human flows at the same site and test whether the simulated spatial population distribution achieves cosine similarity above 0.7 with the observed data.

Figures

Figures reproduced from arXiv: 2605.17426 by Chiharu Shima, Fukuharu Tanaka, Haruki Yonekura, Hirozumi Yamaguchi, Tatsuya Amano.

Figure 1
Figure 1. Figure 1: Predicting the Effect of Mobility Introduction via [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the Proposed Platform. method of Uegaki et al. [9], we introduce a structured move￾ment prior estimated from wide-area location point sequences. Wide-area location data refers to privacy-preserving location￾point sequences collected from mobile devices or location￾based services across a broad spatial region. Such data are typ￾ically anonymized, coarsened in space or time, and sometimes perturb… view at source ↗
Figure 3
Figure 3. Figure 3: Example of Visualized Result. To represent mobility introduction, we do not retrain the decision model on intervention-specific data. Instead, we mod￾ify the environmental features supplied to the trained model. Specifically, for mobility-enabled PoI pairs and affected PoI locations, we replace the effective travel speed with the mo￾bility cruising speed and adjust the attractiveness of associated location… view at source ↗
Figure 6
Figure 6. Figure 6: Each PoI is assigned an attraction score representing the [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mobilities we installed [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Area Partitioning for our Simulation [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: PoI Locations and Mobility Deployment. walking speed to 5.0 km/h. In addition, attraction scores were updated at deployment locations such as location 00, loca￾tion 01, location 03, and location 04. The micromobilities we TABLE II: Population Reproduction Accuracy With Mobility Introduction. Model MAE Day-aggregated MAE MLP 2.34 334 MLP + MoS 2.52 360 GNN 2.87 410 MLP with Exit Class 1.76 252 MLP + MoS wit… view at source ↗
Figure 7
Figure 7. Figure 7: Estimated and Ground Truth Population Distributions With Mobility Introduction. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Grouped SHAP beeswarm plot averaged over classes. [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
read the original abstract

We propose a framework for predicting the effects of mobility introduction measures using a human-flow digital twin. This digital twin incorporates a multi-agent simulator that can represent how visitors choose destinations depending on factors such as their current location and the attractiveness of spots. We extract data on how visitors selected destinations with respect to measured pre-intervention human-flow data, inter-spot distances, spot attractiveness, and travel volumes, and use these data to train each agent's decision model of this simulator. The trained decision model is a function that takes a visitor's current state and surrounding environmental information as input and outputs which spot the visitor will move toward next. By expressing mobility introduction measures as changes to inter-point distances or to spot attractiveness, the framework can reproduce human flows with mobility introduction in the multi-agent simulator and thereby quantify effects such as changes in visitor counts and circulation. We evaluated the proposed method using human-flow data measured with and without introducing mobility within Wakayama Castle Park in Japan. When reproducing flows with mobility introduction using a multi-layer perceptron decision model, the cosine similarity of the spatial population distribution exceeded 0.7, confirming that the approach can replicate the flow changes caused by the mobility introduction.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a human-flow digital twin framework using a multi-agent simulator with an MLP decision model to predict visitor circulation changes from mobility introduction measures. The model is trained on pre-intervention data from Wakayama Castle Park (current location, inter-spot distances, spot attractiveness, travel volumes) to learn destination choice; mobility effects are then simulated by editing those same inputs, with validation reporting cosine similarity >0.7 on the resulting spatial population distribution compared to real post-intervention measurements.

Significance. If the modeling assumptions hold, the framework could enable low-cost prediction of mobility interventions in tourism and public-space settings by reusing a single trained simulator rather than repeated field studies. The grounding in paired real pre/post data from an actual site is a concrete strength that distinguishes it from purely synthetic evaluations.

major comments (2)
  1. [Evaluation] Evaluation section (as summarized in the abstract): the central claim that the framework replicates observed flow changes rests on the untested assumption that editing only inter-spot distances and spot attractiveness inside the trained MLP is sufficient to capture all behavioral effects of the real mobility introduction. No evidence or sensitivity analysis is provided to rule out unmodeled factors (e.g., route awareness or social signaling) that could produce the same aggregate cosine similarity >0.7 without the simulator truly modeling the intervention.
  2. [Methods] Methods section: the manuscript provides no details on the MLP training procedure, feature engineering, train/validation splits, regularization, or statistical significance testing of the cosine similarity result. Without these, it is impossible to determine whether the reported >0.7 similarity reflects robust out-of-sample prediction or site-specific overfitting.
minor comments (2)
  1. [Abstract] The abstract and main text should explicitly state the exact input vector to the MLP and the output encoding (e.g., softmax over spots) so readers can reproduce the decision model.
  2. [Evaluation] Figure captions or the evaluation section should report the number of visitors, time windows, and exact post-intervention modifications applied to distances/attractiveness for full reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment in detail below and outline the revisions we will make to improve clarity and rigor.

read point-by-point responses

  1. Referee: [Evaluation] Evaluation section (as summarized in the abstract): the central claim that the framework replicates observed flow changes rests on the untested assumption that editing only inter-spot distances and spot attractiveness inside the trained MLP is sufficient to capture all behavioral effects of the real mobility introduction. No evidence or sensitivity analysis is provided to rule out unmodeled factors (e.g., route awareness or social signaling) that could produce the same aggregate cosine similarity >0.7 without the simulator truly modeling the intervention.

    Authors: We agree that the evaluation centers on the assumption that mobility effects can be represented primarily through modifications to inter-spot distances and spot attractiveness, which are the variables directly altered in the simulator to model the intervention. The resulting cosine similarity above 0.7 demonstrates that these changes produce aggregate spatial distributions consistent with post-intervention observations. To strengthen the manuscript, we will add a dedicated limitations and robustness subsection that explicitly discusses potential unmodeled behavioral factors such as route awareness and social signaling. We will also incorporate a sensitivity analysis by systematically varying additional simulation parameters (where supported by the available data) and reporting the impact on the similarity metric. revision: yes

  2. Referee: [Methods] Methods section: the manuscript provides no details on the MLP training procedure, feature engineering, train/validation splits, regularization, or statistical significance testing of the cosine similarity result. Without these, it is impossible to determine whether the reported >0.7 similarity reflects robust out-of-sample prediction or site-specific overfitting.

    Authors: We acknowledge the omission of these methodological details, which are necessary for assessing reproducibility and robustness. In the revised manuscript we will expand the Methods section with a new subsection that specifies the MLP architecture (layers, hidden units, activation functions), training procedure (optimizer, learning rate schedule, number of epochs, loss function), feature engineering steps (encoding of location, distances, attractiveness, and volumes), train/validation splits (including ratios and any cross-validation), regularization techniques applied, and statistical evaluation of the cosine similarity (including bootstrap-derived confidence intervals or comparison against a null distribution). revision: yes

Circularity Check

0 steps flagged

No significant circularity; evaluation uses independent post-intervention measurements

full rationale

The derivation trains an MLP decision model exclusively on pre-intervention observations of location, distances, attractiveness, and volumes. Mobility effects are then expressed as explicit input edits to those same quantities, the simulator is run, and the resulting spatial distribution is compared via cosine similarity to separately measured post-intervention human-flow data. Because the evaluation target is external ground-truth data collected after the real intervention, the reported similarity >0.7 does not reduce to the training inputs by construction. No self-citations, uniqueness theorems, or ansatzes are invoked in the provided chain, and the central claim remains falsifiable against the held-out post data.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract-only review limits identification of exact fitted values; the decision model parameters are learned from data rather than chosen ad hoc, but the core modeling assumptions about visitor choice factors are domain-level.

free parameters (1)
  • MLP decision model weights
    The neural network parameters are fitted to pre-intervention visitor choice data to map current state and environment to next destination.
axioms (1)
  • domain assumption Visitor destination choice is determined by current location, inter-spot distances, spot attractiveness, and travel volumes
    This assumption defines the input features for the decision model extracted from measured pre-intervention data.

pith-pipeline@v0.9.0 · 5754 in / 1463 out tokens · 39482 ms · 2026-05-19T22:34:57.541608+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages

  1. [1]

    Shared e-scooter micromobility: review of use patterns, perceptions and environmental impacts,

    H. Badia and E. Jenelius, “Shared e-scooter micromobility: review of use patterns, perceptions and environmental impacts,”Transport Reviews, vol. 43, no. 5, pp. 811–837, 2023

    work page 2023

  2. [2]

    Tourists on shared bikes: Can bike- sharing boost attraction demand?

    Y . Yang, L. Jiang, and Z. Zhang, “Tourists on shared bikes: Can bike- sharing boost attraction demand?”Tourism Management, vol. 86, p. 104328, 2021

    work page 2021

  3. [3]

    Large-scale agent- based simulation model of pedestrian traffic flows,

    D. Kaziyeva, P. Stutz, G. Wallentin, and M. Loidl, “Large-scale agent- based simulation model of pedestrian traffic flows,”Computers, Envi- ronment and Urban Systems, vol. 105, p. 102021, 2023

    work page 2023

  4. [4]

    Integrating discrete choice models with matsim scoring,

    S. H ¨orl, “Integrating discrete choice models with matsim scoring,” Procedia Computer Science, vol. 184, pp. 704–711, 2021

    work page 2021

  5. [5]

    A data-driven framework for the safe integration of micro-mobility into the transport system: Comparing bicycles and e-scooters in field trials,

    M. Dozza, A. Violin, and A. Rasch, “A data-driven framework for the safe integration of micro-mobility into the transport system: Comparing bicycles and e-scooters in field trials,”Journal of safety research, vol. 81, pp. 67–77, 2022

    work page 2022

  6. [6]

    Balancing privacy and planning: using counterfactuals to predict and optimize tourism in wakayama city,

    M. Mishra, S. M. Kala, T. Amano, and H. Yamaguchi, “Balancing privacy and planning: using counterfactuals to predict and optimize tourism in wakayama city,” inProceedings of the 2nd ACM SIGSPA- TIAL international workshop on geo-privacy and data utility for smart societies, 2024, pp. 25–30

    work page 2024

  7. [7]

    A questionnaire- only counterfactual machine learning approach to assess the spatial im- pact of green mobility vehicles in urban parks: A wakayama castle case study,

    R. Naeem, S. Manas, T. Amano, and H. Yamaguchi, “A questionnaire- only counterfactual machine learning approach to assess the spatial im- pact of green mobility vehicles in urban parks: A wakayama castle case study,” inCompanion Proceedings of the 27th International Conference on Distributed Computing and Networking, ser. ICDCN Companion ’26. Association ...

    work page 2026

  8. [8]

    Exploring digital twins for transport planning: a review,

    D. Nag, F. Brandel-Tanis, Z. A. Pramestri, K. Pitera, and Y . K. Frøyen, “Exploring digital twins for transport planning: a review,”European Transport Research Review, vol. 17, no. 1, p. 15, 2025

    work page 2025

  9. [9]

    Simulating urban pedestrian flows by fusing wide-area location data and spot pedestrian counts,

    M. Uegaki, T. Amano, and H. Yamaguchi, “Simulating urban pedestrian flows by fusing wide-area location data and spot pedestrian counts,” in Proceedings of the 21st EAI International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services, 2024

    work page 2024

  10. [10]

    Sumo (simulation of urban mobility); an open-source traffic simulation,

    C. R. Daniel Krajzewicz, Georg Hertkorn and P. Wagner, “Sumo (simulation of urban mobility); an open-source traffic simulation,” in Proceedings of 4th Middle East Symposium on Simulation and Mod- elling, 2002

    work page 2002

  11. [11]

    Jupedsim: an open framework for simulating and analyzing the dynamics of pedestrians,

    J. Z. Armel Ulrich Kemloh Wagoum, Mohcine Chraibi and G. Lam- mel, “Jupedsim: an open framework for simulating and analyzing the dynamics of pedestrians,” inProceedings of the 3rd Conference of Transportation Research Group of India, 2015

    work page 2015

  12. [12]

    Traci: An interface for coupling road traffic and network simulators,

    A. Wegener, M. Piorkowski, M. Raya, H. Hellbr ¨uck, S. Fischer, and J.-P. Hubaux, “Traci: An interface for coupling road traffic and network simulators,” inProceedings of the 11th Communications and Networking Simulation Symposium, April 2008

    work page 2008

  13. [13]

    Specification, estimation and validation of a pedestrian walking behavior model,

    T. Robin, G. Antonini, M. Bierlaire, and J. Cruz, “Specification, estimation and validation of a pedestrian walking behavior model,” Transportation Research Part B: Methodological, vol. 43, no. 1, pp. 36–56, 2009

    work page 2009

  14. [14]

    Towards travel recommendation interpretability: Disentangling tourist decision-making process via knowledge graph,

    J. Gao, P. Peng, F. Lu, C. Claramunt, and Y . Xu, “Towards travel recommendation interpretability: Disentangling tourist decision-making process via knowledge graph,”Information Processing & Management, vol. 60, no. 4, p. 103369, 2023

    work page 2023

  15. [15]

    Travel inspiration in tourist decision making,

    F. Dai, D. Wang, and K. Kirillova, “Travel inspiration in tourist decision making,”Tourism Management, vol. 90, p. 104484, 2022

    work page 2022

  16. [16]

    Time for a radical reappraisal of tourist decision making? toward a new conceptual model,

    S. Mccabe, C. Li, and Z. Chen, “Time for a radical reappraisal of tourist decision making? toward a new conceptual model,”Journal of Travel Research, vol. 55, 07 2015

    work page 2015

  17. [17]

    A virtual experiment on pedestrian desti- nation choice: the role of schedules, the environment and behavioural categories,

    C. King and N. W. Bode, “A virtual experiment on pedestrian desti- nation choice: the role of schedules, the environment and behavioural categories,”Royal Society open science, vol. 9, no. 7, p. 211982, 2022

    work page 2022

  18. [18]

    Exploring the pedestrian route choice behaviors by machine learning models,

    C.-J. Jin, Y . Luo, C. Wu, Y . Song, and D. Li, “Exploring the pedestrian route choice behaviors by machine learning models,”ISPRS Interna- tional Journal of Geo-Information, vol. 13, no. 5, p. 146, 2024

    work page 2024

  19. [19]

    A mathematical model for the behavior of pedestrians,

    D. Helbing, “A mathematical model for the behavior of pedestrians,” Behavioral science, vol. 36, no. 4, pp. 298–310, 1991

    work page 1991

  20. [20]

    Continuum crowds,

    A. Treuille, S. Cooper, and Z. Popovi ´c, “Continuum crowds,”ACM transactions on graphics (TOG), vol. 25, no. 3, pp. 1160–1168, 2006

    work page 2006

  21. [21]

    A speed-based model for crowd simulation considering walking preferences,

    S. Zhang, J. Zhang, M. Chraibi, and W. Song, “A speed-based model for crowd simulation considering walking preferences,”Communications in nonlinear science and numerical simulation, vol. 95, p. 105624, 2021

    work page 2021

  22. [22]

    Hypedsim: A multi- level crowd-simulation framework—methodology, calibration, and vali- dation,

    H.-T. Dang, B. Gaudou, and N. Verstaevel, “Hypedsim: A multi- level crowd-simulation framework—methodology, calibration, and vali- dation,”Sensors, vol. 24, no. 5, p. 1639, 2024

    work page 2024

  23. [23]

    Modeling pedestrian tactical and operational decisions under risk and uncertainty: A two-layer model framework,

    R. Huang, X. Zhao, Y . Yuan, Q. Yu, C. Liu, and W. Daamen, “Modeling pedestrian tactical and operational decisions under risk and uncertainty: A two-layer model framework,”IEEE Transactions on Intelligent Trans- portation Systems, vol. 24, no. 5, pp. 5259–5281, 2023

    work page 2023

  24. [24]

    High-resolution destination choice in agent-based demand models,

    A. Horni, K. Nagel, and K. W. Axhausen, “High-resolution destination choice in agent-based demand models,”Arbeitsberichte Verkehrs-und Raumplanung, vol. 682, 2011

    work page 2011

  25. [25]

    Model- ing intermodal travel behavior in an agent-based travel demand model,

    T. W ¨orle, L. Briem, M. Heilig, M. Kagerbauer, and P. V ortisch, “Model- ing intermodal travel behavior in an agent-based travel demand model,” Procedia Computer Science, vol. 184, pp. 202–209, 2021

    work page 2021

  26. [26]

    A co-simulation system that integrates matsim with a set of external fleet simulations,

    H. Yang, E. Wong, H. Davis III, and J. Y . Chow, “A co-simulation system that integrates matsim with a set of external fleet simulations,” Simulation Modelling Practice and Theory, vol. 134, p. 102957, 2024

    work page 2024

  27. [27]

    Agent-based digital traffic model generation for regions facing data scarcity using aggregated cellphone data: a case study for brussels,

    J. Li, E. Rombaut, and L. Vanhaverbeke, “Agent-based digital traffic model generation for regions facing data scarcity using aggregated cellphone data: a case study for brussels,”International Journal of Digital Earth, vol. 17, no. 1, p. 2407046, 2024

    work page 2024

  28. [28]

    A unified approach to interpreting model predictions,

    S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,”Advances in neural information processing systems, vol. 30, 2017

    work page 2017