pith. sign in

arxiv: 2606.25484 · v1 · pith:CLAUBWLKnew · submitted 2026-06-24 · 💻 cs.CY · econ.GN· q-fin.EC· stat.AP

From Causal Discovery to Implementation: An Agentic AI Framework for E-Scooter Mobility Hub Planning Across 29 German Cities

Meng Jin , Melanie Handrich , Simone Martinenz , Nicholas Hoeser , Ziyue Li This is my paper

Pith reviewed 2026-06-25 19:49 UTC · model grok-4.3

classification 💻 cs.CY econ.GNq-fin.ECstat.AP
keywords causal discoverye-scootermobility hubsurban planningagentic AIGBFS datacity typologiesdemand modeling
0
0 comments X

The pith

A causal template library built from public data across 29 German cities distinguishes core and peripheral drivers of e-scooter demand to inform hub placement.

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

The paper introduces a three-phase agentic AI framework that processes GBFS data from 29 German cities to create a Causal Template Library. This library captures how environmental features causally influence e-scooter hotspot demand differently for core and peripheral clusters within various city types. An LLM-orchestrated pipeline adapts causal discovery algorithms to local conditions across 57 units. The resulting templates support a planning tool that scores candidate sites and generates reports, with evidence of use in actual construction projects in Heilbronn.

Core claim

The framework constructs a Causal Template Library encoding causal drivers of hotspot demand for each combination of city type and cluster type. Core demand is driven by activity access and transit proximity while peripheral demand responds to built form, with city-type-specific patterns enabling transferable siting templates. A planning tool built on the library scores sites, calibrates infrastructure to demographics, and has informed two hub sites now under construction in Heilbronn.

What carries the argument

The Causal Template Library, which encodes city-type and cluster-type specific causal relationships between environmental features and e-scooter hotspot demand.

If this is right

  • Transferable siting templates can be derived for different city types based on the identified causal patterns.
  • The planning tool can score candidate sites and calibrate recommendations using local demographics.
  • Practitioner-ready reports can be generated to support real-world mobility hub decisions.
  • City-type-specific patterns allow for generalizable application across similar urban typologies.

Where Pith is reading between the lines

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

  • Frameworks like this could extend to planning for other micromobility options using similar public data sources.
  • Validation in additional cities beyond Heilbronn would test the transferability of the templates.
  • Integrating demographic data more deeply might further refine the infrastructure recommendations.
  • The approach highlights the potential for public data to replace proprietary trip data in urban planning models.

Load-bearing premise

The causal discovery methods reliably identify true causal effects from the GBFS data without being misled by correlations or unmeasured factors.

What would settle it

If applying the templates to a new set of cities produces demand predictions that do not match observed hotspot locations after controlling for the identified features.

Figures

Figures reproduced from arXiv: 2606.25484 by Melanie Handrich, Meng Jin, Nicholas Hoeser, Simone Martinenz, Ziyue Li.

Figure 1
Figure 1. Figure 1: Study area overview. Figure 1a shows the geographic distribution of the 29 study cities, with point size reflecting population class and colour indicating functional typology. Figure 1b shows normalised mean environmental feature profiles for each functional typology, revealing distinct characteristics that reflect the urban identity of each city type. University cities are distinguished by a high number o… view at source ↗
Figure 2
Figure 2. Figure 2: Spatial validation of GBFS-derived hotspots against concurrent 41-day and full-year MDS data in Heilbronn (300 m [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative environmental feature layers for Stuttgart: (a1) population density at 100 m grid resolution, (a2) dom [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Citizen preference survey results: hub usage rates, location attractiveness, mobility mode importance, and infras [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Three-phase agentic AI framework for e-scooter mobility hub planning. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: OD distance and trip duration distributions for Heilbronn (Apr 2025–Apr 2026). Dashed vertical lines mark the [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Spatial clusters in Heilbronn (k = 2): built-environment z-score profiles (left) and geographic distribution (right). Point size proportional to trip count. 5.1.2. Phase 2: Agentic Causal Discovery Cluster Profiles. The city–cluster profiles for Heilbronn’s two spatial clusters are: Heilbronn (130,093 inhabitants, 99.9 km2 ) is a mid-sized city in northern Baden-Württemberg with predominantly flat terrain … view at source ↗
Figure 8
Figure 8. Figure 8: Consensus causal DAGs for C0 (urban mixed-use, left) and C1 (peripheral, right), Heilbronn. [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Planning tool: 15-minute e-scooter isochrones for Site A (red) and Site B (blue), Heilbronn. [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Site comparison panel: network metrics, POI counts, and mobility usage likelihood for Site A and Site B (15-minute [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Spatial distribution of C0 (core high-activity) and C1 (peripheral low-activity) hotspot clusters for nine representative [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Consensus causal graphs grouped by city size class (columns: large [PITH_FULL_IMAGE:figures/full_fig_p020_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Consensus causal graphs grouped by functional city type (columns: university, port, hilly, industrial) and cluster [PITH_FULL_IMAGE:figures/full_fig_p021_13.png] view at source ↗
read the original abstract

Existing approaches to e-scooter mobility hub planning lack city-type-specific causal evidence. Demand models are typically correlational, built on proprietary trip data, and do not distinguish how driver profiles vary across urban typologies. This paper presents a three-phase agentic AI framework that constructs a Causal Template Library from public GBFS data across 29 German cities, encoding which environmental features causally drive hotspot demand for each combination of city type (large, university, industrial, hilly) and cluster type (core, peripheral). A large language model (LLM) orchestrated causal discovery pipeline adapts algorithm selection to local data conditions across 57 city-cluster units. The library reveals systematic variation. Core demand is driven by activity access and transit proximity, while peripheral demand responds to built form, with city-type-specific patterns supporting transferable siting templates. A planning tool built on the library scores candidate sites, calibrates infrastructure recommendations to local demographics, and generates practitioner-ready reports. In Heilbronn, Germany, two hub sites informed by the framework's causal evidence are currently under construction, illustrating how the outputs can support real-world siting decisions.

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

3 major / 0 minor

Summary. This paper proposes a three-phase agentic AI framework for planning e-scooter mobility hubs. It employs an LLM to orchestrate causal discovery on GBFS data from 29 German cities, building a Causal Template Library that identifies causal drivers of demand differentiated by city type (large, university, industrial, hilly) and core/peripheral clusters. The library supports a planning tool for site scoring and report generation, with two sites in Heilbronn under construction based on its outputs.

Significance. Should the causal relationships prove robust and generalizable beyond the sample, the work would offer a significant advance in mobility hub planning by supplying city-type-specific causal templates derived from public data, moving past purely correlational approaches. The demonstrated real-world implementation in Heilbronn is a positive indicator of practical utility.

major comments (3)
  1. [Abstract] Abstract: The assertion that the LLM-orchestrated pipeline produces 'causal evidence' and a 'Causal Template Library' encoding true causal drivers is load-bearing for all downstream claims, yet the manuscript supplies no validation details, sensitivity analyses, robustness checks, or baseline comparisons for the discovery step on observational GBFS data.
  2. [Causal Template Library construction] Causal Template Library construction: The reported patterns (core demand driven by activity access/transit; peripheral by built form) are derived from the same 57 city-cluster GBFS observations used to fit the pipeline, creating a circularity risk that is not addressed by external benchmarks or pre-registered predictions.
  3. [Heilbronn implementation] Heilbronn implementation: The note that two hub sites are under construction provides downstream consistency but does not validate the causal identification step, as no details are given on how the specific causal edges (versus correlations) informed site selection or any falsification test.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The comments correctly identify that stronger validation of the causal discovery component is needed to support the paper's claims. We respond to each major comment below and commit to revisions that address the identified gaps.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The assertion that the LLM-orchestrated pipeline produces 'causal evidence' and a 'Causal Template Library' encoding true causal drivers is load-bearing for all downstream claims, yet the manuscript supplies no validation details, sensitivity analyses, robustness checks, or baseline comparisons for the discovery step on observational GBFS data.

    Authors: We agree that the manuscript lacks sufficient validation details for the causal discovery step. In the revision we will add a new subsection detailing sensitivity analyses (varying algorithm hyperparameters and data subsamples), robustness checks across multiple causal discovery algorithms, and explicit baseline comparisons against purely correlational models. These additions will directly support the claims of causal drivers. revision: yes

  2. Referee: [Causal Template Library construction] Causal Template Library construction: The reported patterns (core demand driven by activity access/transit; peripheral by built form) are derived from the same 57 city-cluster GBFS observations used to fit the pipeline, creating a circularity risk that is not addressed by external benchmarks or pre-registered predictions.

    Authors: The patterns are data-driven from the 57 observations, which is inherent to constructing templates from the available sample. We will add external benchmark comparisons against independent mobility datasets where feasible and will explicitly discuss the exploratory nature of the analysis. Pre-registration was not performed as this was an exploratory study; we will note this limitation and its implications for generalizability. revision: partial

  3. Referee: [Heilbronn implementation] Heilbronn implementation: The note that two hub sites are under construction provides downstream consistency but does not validate the causal identification step, as no details are given on how the specific causal edges (versus correlations) informed site selection or any falsification test.

    Authors: We will expand the Heilbronn section to specify how particular causal edges (e.g., activity access and transit proximity in core clusters) were used in the site scoring function and how they differed from correlational alternatives in the decision process. We will also describe candidate falsification tests that could be applied once post-implementation data become available. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical causal discovery on observational data is self-contained

full rationale

The paper applies standard causal discovery algorithms (orchestrated by LLM) to public GBFS data across 29 cities to construct a Causal Template Library encoding feature-demand relationships, then uses the resulting library for site scoring and reports. This is a conventional data-driven pipeline with no self-definitional equations, no fitted parameters renamed as out-of-sample predictions, and no load-bearing self-citations or uniqueness theorems invoked. The derivation chain does not reduce any claimed result to its own inputs by construction; the templates are outputs of the discovery step on the same data, which is the intended empirical process rather than a circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the framework implicitly assumes GBFS data quality and LLM reliability for causal tasks but does not enumerate them.

pith-pipeline@v0.9.1-grok · 5755 in / 1329 out tokens · 36556 ms · 2026-06-25T19:49:42.726136+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

56 extracted references · 1 canonical work pages

  1. [1]

    arXiv preprint arXiv:2503.06395 , year=

    Causal Discovery and Inference towards Urban Elements and Associated Factors , author=. arXiv preprint arXiv:2503.06395 , year=

    work page arXiv

  2. [2]

    Journal of Public Transportation , volume=

    Data-driven causal behaviour modelling from trajectory data: A case for fare incentives in public transport , author=. Journal of Public Transportation , volume=. 2025 , publisher=

    2025

  3. [3]

    Proceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems , pages=

    CausalGRIT: Causal Graph Reasoning for Traffic Congestion Prediction--From Statistical Association to Casual Intervention , author=. Proceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems , pages=

  4. [4]

    2025 IEEE 21st International Conference on Automation Science and Engineering (CASE) , pages=

    Data-driven optimization of EV charging station placement using causal discovery , author=. 2025 IEEE 21st International Conference on Automation Science and Engineering (CASE) , pages=. 2025 , organization=

    2025

  5. [5]

    Transportation Research Part D: Transport and Environment , volume=

    Spatial heterogeneity in human mobility responses to London’s ultra-low emission zone expansion , author=. Transportation Research Part D: Transport and Environment , volume=. 2025 , publisher=

    2025

  6. [6]

    Urban Science , volume=

    Urban Mobility and Socio-Environmental Aspects in David, Panama: A Bayesian-Network Analysis , author=. Urban Science , volume=. 2025 , publisher=

    2025

  7. [7]

    Journal of Planning Education and Research , volume=

    The pathway of urban planning AI: From planning support to plan-making , author=. Journal of Planning Education and Research , volume=. 2024 , publisher=

    2024

  8. [8]

    Electronics , volume=

    Artificial intelligence-based decision support system for sustainable urban mobility , author=. Electronics , volume=. 2024 , publisher=

    2024

  9. [9]

    Nature computational science , volume=

    Urban planning in the era of large language models , author=. Nature computational science , volume=. 2025 , publisher=

    2025

  10. [10]

    Urban Informatics , volume=

    Conceptualising the emergence of Agentic Urban AI: from automation to agency , author=. Urban Informatics , volume=. 2025 , publisher=

    2025

  11. [11]

    Transportation Research Record , volume=

    Micromobility trip origin and destination inference using general bikeshare feed specification data , author=. Transportation Research Record , volume=. 2022 , publisher=

    2022

  12. [12]

    Journal of Statistical Computation and Simulation , volume=

    Shapiro--Wilk test for skew normal distributions based on data transformations , author=. Journal of Statistical Computation and Simulation , volume=. 2019 , publisher=

    2019

  13. [13]

    Earth System Science Data , volume =

    Zhu, Xiao Xiang and Chen, Songhao and Zhang, Fahong and Shi, Yilei and Wang, Yuanyuan , title =. Earth System Science Data , volume =. 2025 , doi =

    2025

  14. [14]

    Proceedings of the 5th International Conference on Applications of Digital Information and Web Technologies (ICADIWT) , pages =

    Khan, Kamran and Rehman, Saif Ur and Aziz, Kamran and Fong, Simon and Sarasvady, Sababady , title =. Proceedings of the 5th International Conference on Applications of Digital Information and Web Technologies (ICADIWT) , pages =

  15. [15]

    Spirtes, Peter and Glymour, Clark and Scheines, Richard , title =

  16. [16]

    , author=

    Estimating high-dimensional directed acyclic graphs with the PC-algorithm. , author=. Journal of Machine Learning Research , volume=

  17. [17]

    , author=

    A linear non-Gaussian acyclic model for causal discovery. , author=. Journal of Machine Learning Research , volume=

  18. [18]

    Journal of Machine Learning Research , volume =

    Shimizu, Shohei and Inazumi, Takanori and Sogawa, Yasuhiro and Hyv. Journal of Machine Learning Research , volume =

  19. [19]

    , title =

    Zheng, Xun and Aragam, Bryon and Ravikumar, Pradeep and Xing, Eric P. , title =. Advances in Neural Information Processing Systems (NeurIPS) , volume =

  20. [20]

    Advances in Neural Information Processing Systems , volume=

    Dagma: Learning dags via m-matrices and a log-determinant acyclicity characterization , author=. Advances in Neural Information Processing Systems , volume=

  21. [21]

    Journal of machine learning research , volume=

    Optimal structure identification with greedy search , author=. Journal of machine learning research , volume=

  22. [22]

    Quality & quantity , volume=

    A caution regarding rules of thumb for variance inflation factors , author=. Quality & quantity , volume=. 2007 , publisher=

    2007

  23. [23]

    Machine learning , volume=

    Random forests , author=. Machine learning , volume=. 2001 , publisher=

    2001

  24. [24]

    Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , volume=

    Estimating mutual information , author=. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , volume=. 2004 , publisher=

    2004

  25. [25]

    European journal of operational research , volume=

    Exploratory data analysis , author=. European journal of operational research , volume=. 1986 , publisher=

    1986

  26. [26]

    Journal of machine learning research , volume=

    Joint causal inference from multiple contexts , author=. Journal of machine learning research , volume=

  27. [27]

    Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages=

    Xgboost: A scalable tree boosting system , author=. Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages=

  28. [28]

    Bmj , volume=

    Pearson’s correlation coefficient , author=. Bmj , volume=. 2012 , publisher=

    2012

  29. [29]

    2017 , publisher=

    Elements of causal inference: foundations and learning algorithms , author=. 2017 , publisher=

    2017

  30. [30]

    Box, George E. P. and Cox, David R. , title =. Journal of the Royal Statistical Society, Series B , volume =

  31. [31]

    Journal of the Royal Statistical Society, Series B , volume =

    Zou, Hui and Hastie, Trevor , title =. Journal of the Royal Statistical Society, Series B , volume =

  32. [32]

    Monographs on statistics and applied probability , volume=

    An introduction to the bootstrap , author=. Monographs on statistics and applied probability , volume=

  33. [33]

    Annual review of statistics and its application , volume=

    High-dimensional statistics with a view toward applications in biology , author=. Annual review of statistics and its application , volume=. 2014 , publisher=

    2014

  34. [34]

    Perspectives in Electronic Structure Theory , pages=

    Elements of information theory , author=. Perspectives in Electronic Structure Theory , pages=. 2011 , publisher=

    2011

  35. [35]

    Uncertainty in Artificial Intelligence , pages=

    Greedy relaxations of the sparsest permutation algorithm , author=. Uncertainty in Artificial Intelligence , pages=. 2022 , organization=

    2022

  36. [36]

    Course lecture notes (draft) , volume=

    Introduction to causal inference , author=. Course lecture notes (draft) , volume=

  37. [37]

    Computers, environment and urban systems , volume=

    OSMnx: New methods for acquiring, constructing, analyzing, and visualizing complex street networks , author=. Computers, environment and urban systems , volume=. 2017 , publisher=

    2017

  38. [38]

    Transportation research part D: transport and environment , volume=

    Understanding factors influencing shared e-scooter usage and its impact on auto mode substitution , author=. Transportation research part D: transport and environment , volume=. 2021 , publisher=

    2021

  39. [39]

    Transport reviews , volume=

    Shared e-scooter micromobility: review of use patterns, perceptions and environmental impacts , author=. Transport reviews , volume=. 2023 , publisher=

    2023

  40. [40]

    Transportation Research Part D: Transport and Environment , volume=

    Spatial associations of dockless shared e-scooter usage , author=. Transportation Research Part D: Transport and Environment , volume=. 2020 , publisher=

    2020

  41. [41]

    Journal of transport geography , volume=

    Influence of the built environment on E-scooter sharing ridership: A tale of five cities , author=. Journal of transport geography , volume=. 2021 , publisher=

    2021

  42. [42]

    Journal of transport geography , volume=

    Spatiotemporal comparative analysis of scooter-share and bike-share usage patterns in Washington, DC , author=. Journal of transport geography , volume=. 2019 , publisher=

    2019

  43. [43]

    Transportation Research Part D: Transport and Environment , volume=

    Comprehensive comparison of e-scooter sharing mobility: Evidence from 30 European cities , author=. Transportation Research Part D: Transport and Environment , volume=. 2022 , publisher=

    2022

  44. [44]

    Transportation research part C: emerging technologies , volume=

    Integrating shared e-scooters as the feeder to public transit: A comparative analysis of 124 European cities , author=. Transportation research part C: emerging technologies , volume=. 2024 , publisher=

    2024

  45. [45]

    Transportation Research Part D: Transport and Environment , volume=

    Exploring spatial heterogeneity of e-scooter’s relationship with ridesourcing using explainable machine learning , author=. Transportation Research Part D: Transport and Environment , volume=. 2024 , publisher=

    2024

  46. [46]

    Case Studies on Transport Policy , volume=

    Understanding the uneven use of rental e-scooters and implications for equity: Evidence from England’s largest e-scooter trial , author=. Case Studies on Transport Policy , volume=. 2025 , publisher=

    2025

  47. [47]

    Sustainable Cities and Society , volume=

    The effect of shared e-scooter programs on modal shift: Evidence from Sweden , author=. Sustainable Cities and Society , volume=. 2024 , publisher=

    2024

  48. [48]

    Travel behaviour and society , volume=

    Dockless E-scooter usage patterns and urban built Environments: A comparison study of Austin, TX, and Minneapolis, MN , author=. Travel behaviour and society , volume=. 2020 , publisher=

    2020

  49. [49]

    Transportation research procedia , volume=

    Evolution and characteristics of shared e-scooters usage in Munich, Germany--Results of an over 8 million trips data analysis , author=. Transportation research procedia , volume=. 2024 , publisher=

    2024

  50. [50]

    Case Studies on Transport Policy , volume=

    Braving the elements: A time series analysis of e-scooter ridership assessing the impact of weather and seasonality across different climate regions , author=. Case Studies on Transport Policy , volume=. 2025 , publisher=

    2025

  51. [51]

    Transportation Research Part D: Transport and Environment , volume=

    Built environment impacts on zonal shared e-scooter expenses: A Bayesian learning approach , author=. Transportation Research Part D: Transport and Environment , volume=. 2025 , publisher=

    2025

  52. [52]

    Transportation Research Part A: Policy and Practice , volume=

    Shared e-scooter parking regulation: Effects on rider attitudes, perceptions, and use , author=. Transportation Research Part A: Policy and Practice , volume=. 2025 , publisher=

    2025

  53. [53]

    Multimodal Transportation , volume=

    Exploring shared e-scooter trip patterns and links to public transport service level , author=. Multimodal Transportation , volume=. 2025 , publisher=

    2025

  54. [54]

    2019 , note =

    Mobility Data Specification (. 2019 , note =

    2019

  55. [55]

    Advances in Neural Information Processing Systems , volume=

    TraffiDent: A Dataset for Understanding the Interplay between Traffic Dynamics and Incidents , author=. Advances in Neural Information Processing Systems , volume=

  56. [56]

    Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

    Mm-dag: Multi-task dag learning for multi-modal data-with application for traffic congestion analysis , author=. Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=