pith. sign in

arxiv: 2605.31210 · v1 · pith:CWCPCVEEnew · submitted 2026-05-29 · 💻 cs.RO · cs.AI

Simulation of collision avoidance behavior in crowd movement by data-driven approach

Xuanwen Liang , Eric Wai Ming Lee This is my paper

Pith reviewed 2026-06-28 22:04 UTC · model grok-4.3

classification 💻 cs.RO cs.AI
keywords crowd simulationcollision avoidancegenerative adversarial networkbidirectional flowdata-driven modelpedestrian dynamicslateral acceleration
0
0 comments X

The pith

A lateral-acceleration collision penalty added to GAN training reduces opposite-direction collisions in simulated bidirectional crowds to match controlled experiments.

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

The paper develops CPGAN, a generative adversarial network for pedestrian trajectory simulation that adds a new loss term based on lateral acceleration to discourage collision-prone movements. It also uses Voronoi diagrams to extract local motion features from training data. The approach is tested specifically on bidirectional flow scenarios where collisions are frequent in standard data-driven models. A reader would care because high collision rates have limited the usefulness of learned crowd models for safety and layout planning, and this method directly embeds an avoidance mechanism into the training objective instead of relying on post-processing.

Core claim

The central claim is that the lateral-acceleration-based collision loss, when incorporated into the GAN objective, produces simulated trajectories whose opposite-direction collision rates fall to levels observed in controlled experiments, while the generated flows also form lanes and match measured N-t curves.

What carries the argument

The lateral-acceleration-based collision loss function, which quantifies and penalizes predicted sideways movements that would cause overlaps between pedestrians and is added to the standard adversarial loss.

If this is right

  • Bidirectional crowd simulations now exhibit realistic lane formation without excessive head-on overlaps.
  • N-t curves in the simulated flows align with those recorded in real pedestrian experiments.
  • The same penalty approach offers a template for embedding other pedestrian dynamics rules directly into data-driven loss functions.
  • Collision rates in opposite-direction interactions become comparable to those measured under controlled conditions.

Where Pith is reading between the lines

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

  • The method could be tested on multidirectional flows by extending the loss to account for more interaction angles.
  • Facility designers might run the model on proposed layouts to estimate avoidance success before physical construction.
  • Retraining on datasets with varying densities would check whether the learned avoidance generalizes beyond the original training conditions.

Load-bearing premise

That the collision penalty will cause the generator to learn general avoidance behaviors that transfer to new crowd scenarios rather than merely reproducing the exact paths seen during training.

What would settle it

Training the model on one set of bidirectional trajectories and then measuring collision rates on a fresh set of bidirectional flows with different densities or widths, expecting rates to stay near experimental levels.

Figures

Figures reproduced from arXiv: 2605.31210 by Eric Wai Ming Lee, Xuanwen Liang.

Figure 1
Figure 1. Figure 1: Schematic of the controlled experimental setup (refer to https://ped.fz [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Collision avoidance. (a) Snapshot of the controlled experiment [36]. (b) Dyadic collision-avoidance [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Time evolution of lateral acceleration over the avoidance phase. Here, for each pedestrian, the lateral [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the CPGAN model. 3.2. Data processing (DP) module The DP module is responsible for extracting features that govern pedestrian motion, pri￾marily arising from the pedestrian’s and neighbors’ motion as well as the intended destination [12, 4]. To effectively capture these features, we propose a Voronoi-based motion feature ex￾traction approach. Four features are extracted: density ρ (i,t) , x-vel… view at source ↗
Figure 5
Figure 5. Figure 5: Schematic of the Voronoi-based feature extraction. (a) Voronoi diagram in a corridor at time step [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Structure of the VP module. (a) Architecture of the convolutional block. Conv2d ( [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Pedestrian trajectories and time-aligned snapshot distributions from controlled experiments and simula [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: presents the N-t curves from the controlled experiments and simulations. It can be observed that our model effectively reproduces the N-t dynamics, as the simulated curves are in good agreement with those from the controlled experiments [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
read the original abstract

Crowd movement simulation is essential for pedestrian safety management and facility layout optimization. Data-driven models enhance trajectory prediction accuracy under Euclidean metrics, yet they suffer from excessively high collision rates, especially in bidirectional and multidirectional flows. In this paper, we establish a novel data-driven crowd simulation model that incorporates the pedestrian collision mechanism into the loss function to reduce collisions. A new lateral-acceleration-based collision loss function and a Voronoi-based motion feature extraction approach are proposed. The model is based on a Generative Adversarial Network (GAN) architecture and is termed CPGAN (Collision-Penalized GAN). We evaluate CPGAN in bidirectional flow scenarios, which involve frequent collision avoidance behaviors. Results show that the proposed lateral-acceleration-based collision loss significantly reduces opposite-direction pedestrian collision rates to levels comparable with controlled experiments. CPGAN effectively simulates bidirectional flow, reproducing lane formation and N-t curves. The research outcomes can provide inspiration for integrating pedestrian dynamics mechanisms into loss functions in data-driven crowd simulation.

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. The manuscript proposes CPGAN, a GAN-based data-driven model for crowd simulation that incorporates a novel lateral-acceleration-based collision loss term and Voronoi-based motion feature extraction. It evaluates the approach in bidirectional flow scenarios and claims that the collision penalty reduces opposite-direction pedestrian collision rates to levels comparable with controlled experiments while reproducing lane formation and N-t curves.

Significance. If the generalization claims hold under proper validation, integrating a collision-avoidance mechanism directly into the training loss could offer a practical way to improve physical realism in data-driven pedestrian models without sacrificing predictive accuracy on Euclidean metrics. This would be relevant for applications in safety management and facility optimization.

major comments (3)
  1. [Abstract] Abstract: The central claim that the lateral-acceleration-based collision loss 'significantly reduces opposite-direction pedestrian collision rates to levels comparable with controlled experiments' is presented without any reported quantitative metrics, baselines, error bars, data handling procedures, or statistical comparisons, rendering the result unverifiable from the given text.
  2. [Evaluation] Evaluation section: No details are provided on train/test splits or on whether test scenarios differ from training data in key parameters such as pedestrian density, flow ratio, or entry angles; without this, it is impossible to determine whether the observed collision reduction reflects a transferable avoidance mechanism or simply reproduction of low-collision patterns already present in the training trajectories.
  3. [Results] Results: The manuscript reports reproduction of lane formation and N-t curves but contains no ablation experiments that isolate the contribution of the proposed collision loss from the Voronoi features or the base GAN architecture, leaving the load-bearing role of the new loss term unsubstantiated.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight areas where additional clarity and evidence will strengthen the manuscript. We respond to each major comment below and will incorporate revisions as indicated.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the lateral-acceleration-based collision loss 'significantly reduces opposite-direction pedestrian collision rates to levels comparable with controlled experiments' is presented without any reported quantitative metrics, baselines, error bars, data handling procedures, or statistical comparisons, rendering the result unverifiable from the given text.

    Authors: We agree that the abstract should include quantitative support for the claim to improve verifiability. The results section of the manuscript reports collision rate comparisons (including to experimental baselines), but these details are not summarized in the abstract. We will revise the abstract to include key quantitative metrics, such as the reported collision rate reductions and their comparability to controlled experiments. revision: yes

  2. Referee: [Evaluation] Evaluation section: No details are provided on train/test splits or on whether test scenarios differ from training data in key parameters such as pedestrian density, flow ratio, or entry angles; without this, it is impossible to determine whether the observed collision reduction reflects a transferable avoidance mechanism or simply reproduction of low-collision patterns already present in the training trajectories.

    Authors: We acknowledge that explicit information on data partitioning and scenario variation is necessary to assess transferability. The manuscript describes the bidirectional flow evaluation scenarios, but we will add a dedicated subsection detailing the train/test split methodology, along with the specific ranges of pedestrian density, flow ratio, and entry angles used in testing versus training to demonstrate that the collision reduction is not merely memorization. revision: yes

  3. Referee: [Results] Results: The manuscript reports reproduction of lane formation and N-t curves but contains no ablation experiments that isolate the contribution of the proposed collision loss from the Voronoi features or the base GAN architecture, leaving the load-bearing role of the new loss term unsubstantiated.

    Authors: We recognize that ablation studies are required to isolate the contribution of the lateral-acceleration collision loss. The current results compare the full CPGAN model against other approaches, but we will add ablation experiments in the revised manuscript that systematically remove or vary the collision loss term (while retaining Voronoi features and the GAN architecture) to quantify its specific impact on collision rates and trajectory realism. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper introduces an independently motivated lateral-acceleration-based collision loss term added to a standard GAN objective, together with Voronoi feature extraction. The central result is an empirical reduction in collision rates when evaluated against external controlled experiments. No equations or claims in the abstract reduce a prediction to a fitted parameter by construction, invoke self-citations as load-bearing uniqueness theorems, or smuggle ansatzes via prior work. The derivation remains data-driven against held-out or experimental benchmarks rather than tautological.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only view limits visibility into parameters; the collision loss weight and any scaling constants in the lateral-acceleration term are likely tuned during training but not enumerated.

axioms (1)
  • domain assumption Minimizing the lateral-acceleration collision loss during GAN training produces trajectories whose avoidance behavior matches real pedestrians
    The paper relies on this to claim the loss improves realism.

pith-pipeline@v0.9.1-grok · 5697 in / 1148 out tokens · 26957 ms · 2026-06-28T22:04:46.954460+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

45 extracted references · 4 canonical work pages · 2 internal anchors

  1. [1]

    Helbing, I

    D. Helbing, I. Farkas, T. Vicsek, Simulating dynamical features of escape panic, Nature 407 (6803) (2000) 487–490

    2000

  2. [2]

    Varas, M

    A. Varas, M. Cornejo, D. Mainemer, B. Toledo, J. Rogan, V . Muñoz, J. Valdivia, Cellular automaton model for evacuation process with obstacles, Physica A: Statistical Mechanics and its Applications 382 (2) (2007) 631–642

    2007

  3. [3]

    X. Zhao, L. Xia, J. Zhang, W. Song, Artificial neural network based modeling on uni- directional and bidirectional pedestrian flow at straight corridors, Physica A: Statistical Mechanics and its Applications 547 (2020) 123825. 16 Algorithm 1Expected Minimum Distance Calculation via Time Discretization Require:State of pedestriani: positionp (i,t) = [p(i,t...

    2020

  4. [5]

    W. Xie, N. Jiang, Y . Ma, E. Wai Ming Lee, X. Li, H. Yu, Simulating pedestrian flow on slopes via transfer learning approach: From single-file to crowd, IEEE Transactions on Intelligent Transportation Systems 26 (3) (2025) 3873–3884

    2025

  5. [6]

    Jiang, L

    N. Jiang, L. Yang, R. K. K. Yuen, C. Zhai, Modeling the pedestrian flow before bottleneck through learning-based method, IEEE Transactions on Intelligent Transportation Systems 24 (7) (2023) 7035–7047

    2023

  6. [7]

    Jiang, E

    N. Jiang, E. W. M. Lee, L. Yang, R. K. K. Yuen, C. Zhai, A deep-learning-based ap- proach for simulating pedestrian turning flow, Expert Systems with Applications 262 (2025) 125706

    2025

  7. [8]

    H. Li, Z. Liu, B. Zhou, Modeling analysis of t-shaped crowd flow based on artificial neural network, in: CIBDA 2022; 3rd International Conference on Computer Information and Big Data Applications, 2022, pp. 1–5

    2022

  8. [9]

    T. Wang, Z. Zhang, T. Nong, W. Zhang, Y . Tian, Y . Ma, E. W. M. Lee, M. Shi, Simulating pedestrian movement in t-junction corridor: A novel vision-driven convolutional graph attention model with a dataset from experiments, Physica A: Statistical Mechanics and its Applications 674 (2025) 130775

    2025

  9. [12]

    X. Song, K. Chen, X. Li, J. Sun, B. Hou, Y . Cui, B. Zhang, G. Xiong, Z. Wang, Pedestrian trajectory prediction based on deep convolutional lstm network, IEEE Transactions on Intelligent Transportation Systems 22 (6) (2021) 3285–3302

    2021

  10. [13]

    Liang, E

    X. Liang, E. W. M. Lee, Visual-information-driven model for crowd simulation using temporal convolutional network, IEEE Transactions on Intelligent Transportation Systems 25 (9) (2024) 12297–12314

    2024

  11. [14]

    Liang, J

    X. Liang, J. Chen, E. W. M. Lee, W. Xie, Improved visual-information-driven model for crowd simulation and its modular application, Chaos, Solitons & Fractals 209 (2026) 118481

    2026

  12. [15]

    Korbmacher, H.-T

    R. Korbmacher, H.-T. Dang, A. Tordeux, B. Gaudou, N. Verstaevel, Differences in pedes- trian trajectory predictions for high- and low-density situations, in: 14th International Conference on Traffic and Granular Flow (TGF 2022), Springer, Dehli, India, 2022. 18

    2022

  13. [16]

    Alahi, K

    A. Alahi, K. Goel, V . Ramanathan, A. Robicquet, L. Fei-Fei, S. Savarese, Social lstm: Human trajectory prediction in crowded spaces, IEEE, 2016, p. 961–971

    2016

  14. [17]

    Gupta, J

    A. Gupta, J. Johnson, L. Fei-Fei, S. Savarese, A. Alahi, Social gan: Socially acceptable trajectories with generative adversarial networks, 2018 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (2018) 2255–2264

    2018

  15. [18]

    H.-T. Dang, R. Korbmacher, A. Tordeux, B. Gaudou, N. Verstaevel, Ttc-slstm: Human trajectory prediction using time-to-collision interaction energy, in: 2023 15th International Conference on Knowledge and Systems Engineering (KSE), 2023, pp. 1–6

    2023

  16. [19]

    K. Xu, Z. Qin, G. Wang, K. Huang, S. Ye, H. Zhang, Collision-free lstm for human trajec- tory prediction, in: K. Schoeffmann, T. H. Chalidabhongse, C. W. Ngo, S. Aramvith, N. E. O’Connor, Y .-S. Ho, M. Gabbouj, A. Elgammal (Eds.), MultiMedia Modeling, Springer International Publishing, Cham, 2018, pp. 106–116

    2018

  17. [20]

    J. Yue, D. Manocha, H. Wang, Human trajectory prediction via neural social physics, in: S. Avidan, G. Brostow, M. Cissé, G. M. Farinella, T. Hassner (Eds.), Computer Vision – ECCV 2022, Springer Nature Switzerland, Cham, 2022, pp. 376–394

    2022

  18. [21]

    J. Yue, D. Manocha, H. Wang, Human trajectory forecasting with explainable behavioral uncertainty, ArXiv abs/2307.01817 (2023)

    work page arXiv 2023

  19. [22]

    Y . Ma, E. W. M. Lee, R. K. K. Yuen, An artificial intelligence-based approach for sim- ulating pedestrian movement, IEEE Transactions on Intelligent Transportation Systems 17 (11) (2016) 3159–3170

    2016

  20. [23]

    Y . Ma, E. W. Lee, Z. Hu, M. Shi, R. K. Yuen, An intelligence-based approach for predic- tion of microscopic pedestrian walking behavior, IEEE Transactions on Intelligent Trans- portation Systems 20 (10) (2019) 3964–3980

    2019

  21. [24]

    X. Zhao, L. Xia, J. Zhang, W. Song, Artificial neural network based modeling on uni- directional and bidirectional pedestrian flow at straight corridors, Physica A: Statistical Mechanics and its Applications 547 (2020) 123825

    2020

  22. [25]

    X. Zhao, J. Zhang, W. Song, A radar-nearest-neighbor based data-driven approach for crowd simulation, Transportation Research Part C: Emerging Technologies 129 (2021) 103260

    2021

  23. [26]

    Zhang, J

    B. Zhang, J. Xu, C.-Z. T. Xie, S. Lo, B. Zhu, T.-Q. Tang, Y . Tian, Forecaster as a sim- ulator: Simulating multi-directional pedestrian flow with knowledge-guided graph neural networks, Computers & Industrial Engineering 198 (2024) 110668

    2024

  24. [27]

    D. R. Parisi, P. A. Negri, L. Bruno, Experimental characterization of collision avoidance in pedestrian dynamics, Phys. Rev. E 94 (2016) 022318

    2016

  25. [28]

    Y . Qu, J. Wu, H. Yin, X. Yang, Y . Xiao, Analyzing pedestrian individual and interaction collision avoidance dynamics in traditional scenarios, Transportation Research Part C: Emerging Technologies 133 (2021) 103445. 19

    2021

  26. [29]

    Q. Y . Luan, S. B. Liu, Z. J. Fu, J. Y . Lyu, Experimental and modelling studies of collision avoidance strategy choices and behavioural characteristics in interweaving pedestrian flow, Royal Society Open Science 9 (7) (2022) 220187

    2022

  27. [31]

    L. Fu, Y . Liu, H. Qin, Q. Shi, Y . Zhang, Y . Shi, J. T. Lo, An experimental study on bidirectional pedestrian flow involving individuals with simulated disabilities in a corridor, Safety Science 150 (2022) 105723

    2022

  28. [32]

    C.-J. Jin, R. Jiang, S. Wong, S. Xie, D. Li, N. Guo, W. Wang, Observational character- istics of pedestrian flows under high-density conditions based on controlled experiments, Transportation Research Part C: Emerging Technologies 109 (2019) 137–154

    2019

  29. [33]

    Zanlungo, C

    F. Zanlungo, C. Feliciani, Z. Yücel, K. Nishinari, T. Kanda, Macroscopic and micro- scopic dynamics of a pedestrian cross-flow: Part ii, modelling, Safety Science 158 (2023) 105969

    2023

  30. [34]

    Zanlungo, C

    F. Zanlungo, C. Feliciani, Z. Yücel, K. Nishinari, T. Kanda, Macroscopic and microscopic dynamics of a pedestrian cross-flow: Part i, experimental analysis, Safety Science 158 (2023) 105953

    2023

  31. [35]

    Huber, Y .-H

    M. Huber, Y .-H. Su, M. Krüger, K. Faschian, S. Glasauer, J. Hermsdörfer, Adjustments of speed and path when avoiding collisions with another pedestrian, PLOS ONE 9 (2) (2014) 1–13

    2014

  32. [36]

    Zhang, W

    J. Zhang, W. Klingsch, A. Schadschneider, A. Seyfried, Ordering in bidirectional pedes- trian flows and its influence on the fundamental diagram, Journal of Statistical Mechanics: Theory and Experiment 2012 (02) (2012) P02002

    2012

  33. [37]

    W. Xie, E. W. M. Lee, Y . Y . Lee, Self-organisation phenomena in pedestrian counter flows and its modelling, Safety Science 155 (2022) 105875

    2022

  34. [38]

    C. Xuan, D. Xu, N. Jiang, E. W. M. Lee, W. Xie, Unveiling the dynamics and decision- making of dyadic head-on pedestrian collision avoidance: An empirical study, Chaos, Solitons & Fractals 208 (2026) 118126

    2026

  35. [39]

    Steffen, A

    B. Steffen, A. Seyfried, Methods for measuring pedestrian density, flow, speed and direc- tion with minimal scatter, Physica A: Statistical Mechanics and its Applications 389 (9) (2010) 1902–1910

    2010

  36. [40]

    Zhang, W

    J. Zhang, W. Klingsch, A. Schadschneider, A. Seyfried, Transitions in pedestrian funda- mental diagrams of straight corridors and t-junctions, Journal of Statistical Mechanics: Theory and Experiment 2011 (06) (2011) P06004. 20

    2011

  37. [41]

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, Y . Bengio, Generative adversarial networks (2014).arXiv:1406.2661. URLhttps://arxiv.org/abs/1406.2661

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  38. [42]

    V . Nair, G. E. Hinton, Rectified linear units improve restricted boltzmann machines, in: Icml, 2010

    2010

  39. [43]

    Srivastava, G

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, R. Salakhutdinov, Dropout: a simple way to prevent neural networks from overfitting, The journal of machine learning research 15 (1) (2014) 1929–1958

    2014

  40. [44]

    E. T. Hall, R. L. Birdwhistell, B. Bock, P. Bohannan, A. R. Diebold, M. Durbin, M. S. Edmonson, J. L. Fischer, D. Hymes, S. T. Kimball, W. La Barre, , J. E. McClellan, D. S. Marshall, G. B. Milner, H. B. Sarles, G. L. Trager, A. P. Vayda, Proxemics [and comments and replies], Current Anthropology 9 (2/3) (1968) 83–108.doi:10.1086/200975

    work page doi:10.1086/200975 1968

  41. [45]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization (2017).arXiv: 1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  42. [46]

    Hoogendoorn, W

    S. Hoogendoorn, W. Daamen, Self-organization in pedestrian flow, in: S. P. Hoogendoorn, S. Luding, P. H. L. Bovy, M. Schreckenberg, D. E. Wolf (Eds.), Traffic and Granular Flow ’03, Springer Berlin Heidelberg, Berlin, Heidelberg, 2005, pp. 373–382

    2005

  43. [47]

    M. Rex, H. Löwen, Lane formation in oppositely charged colloids driven by an electric field: Chaining and two-dimensional crystallization, Phys. Rev. E 75 (2007) 051402

    2007

  44. [48]

    Nowak, A

    S. Nowak, A. Schadschneider, Quantitative analysis of pedestrian counterflow in a cellular automaton model, Phys. Rev. E 85 (2012) 066128

    2012

  45. [49]

    Feliciani, K

    C. Feliciani, K. Nishinari, Empirical analysis of the lane formation process in bidirectional pedestrian flow, Phys. Rev. E 94 (2016) 032304. 21

    2016