Pith. sign in

REVIEW 3 major objections 6 minor 27 references

The paper claims that interactive trajectory planning for autonomous driving can be cast as one unified potential field, built by fusing a benefit field and a risk field through a physics-inspired variational equation, and solved as a mean-

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 21:10 UTC pith:52IAK2UR

load-bearing objection A creative fusion of MFG and potential fields whose central convergence proof is asserted rather than derived; the simulations don't fill the gap. the 3 major comments →

arxiv 2509.08147 v1 pith:52IAK2UR submitted 2025-09-09 cs.RO

Mean Field Game-Based Interactive Trajectory Planning Using Physics-Inspired Unified Potential Fields

Zhen Tian , Fujiang Yuan , Chunhong Yuan , Yanhong Peng This is my paper
classification cs.RO MSC 91A1693E2049N90
keywords autonomous drivingtrajectory planningmean field gamespotential fieldsdriving style heterogeneityCahn-Hilliard dynamicsstochastic optimal controlNash equilibrium
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper proposes a trajectory-planning framework for autonomous driving called IUPF, which folds two complementary maps of the road into a single unified potential field: a benefit field that attracts the vehicle toward efficient opportunities, and a risk field that repels it from collisions. The fusion is modeled with a Cahn-Hilliard-style variational equation, and the resulting field is used inside a mean-field game in which each vehicle minimizes a stochastic optimal-control cost. The central claim is that this single unified field, with no external safety critic, captures conservative, aggressive, and cooperative driver behavior, and that the resulting best-response iteration has a unique Nash equilibrium with exponential convergence. If the claim holds, interactive planning in dense traffic would need only one tunable field representation instead of layered safety modules, while retaining game-theoretic consistency and scalability.

Core claim

The central claim is that safety and interaction in multi-vehicle planning can be represented by one physics-based scalar field, the Interaction-Enriched Unified Potential Field, which is the steady state of a modified Cahn-Hilliard equation driven by a benefit field and a risk field. Each vehicle solves a forward-backward stochastic differential equation system whose running cost rewards high values of this unified field; the population distribution and the optimal controls are coupled through a mean-field game. The paper states Theorems 1 and 2 asserting that, under convexity and Lipschitz assumptions, the best-response map is a contraction, so a unique Nash equilibrium exists and the iter

What carries the argument

The central object is the unified potential field Phi, obtained by running a modified Cahn-Hilliard equation to steady state. The equation separates the plane into benefit-dominated and risk-dominated regions; its coupling term combines normalized benefit B and risk R with nonlinear and gradient terms. The steady state Phi* then enters the running cost of each vehicle's stochastic optimal-control problem. The argument that this yields a game equilibrium is carried by the best-response operator Gamma on the space of probability measures with the 2-Wasserstein metric; the paper claims Gamma is a contraction whose fixed point is the Nash equilibrium and whose iterates converge exponentially.

Load-bearing premise

The proof assumes that repeatedly recomputing each vehicle's optimal response to the current traffic distribution shrinks the distance between any two distributions, and that the parameters can be chosen to make that shrinkage strict; the paper asserts this but never derives the needed bound.

What would settle it

Numerically evaluate the best-response operator in the paper's lane-change simulation by computing optimal controls for two different initial population distributions and measuring the 2-Wasserstein distance between the resulting trajectory laws. If the distance does not shrink at every step by a factor strictly less than 1, or if two initial distributions converge to different fixed points, the existence-uniqueness and exponential-convergence claims are false for that setting.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • A single potential field can encode both 'go' and 'no-go' tendencies, so trajectory planning does not need a separate safety monitor such as TTC checks or control barrier certificates.
  • Because driving style enters only through the field parameters, aggressive, conservative, and cooperative vehicles are planned by the same equations rather than by different controllers.
  • The mean-field formulation gives a population-level description, so the method should scale to many surrounding vehicles without enumerating pairwise games.
  • The exponential-convergence claim implies a fixed-point solver can be run online in receding-horizon planning with predictable iteration counts.
  • In the reported simulations, the resulting trajectories keep the minimum distance above the 8 m threshold and complete lane-change and overtaking maneuvers smoothly.

Where Pith is reading between the lines

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

  • Not tested in the paper: because the style parameter theta is continuous, the same field construction could in principle infer a driver's aggressiveness online from observed trajectories, turning planning into a joint estimation-and-control problem.
  • The paper validates on only four vehicles, but the mean-field approximation is designed for large populations; whether the Nash-equilibrium guarantee survives at realistic highway densities is an open scaling question.
  • The Cahn-Hilliard interface width and coupling coefficients are hand-tuned; an automatic calibration, for example by matching human driving demonstrations, would be a natural next step.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes IUPF, an interactive trajectory planning framework for autonomous driving that combines mean field game theory with style-dependent benefit and risk potential fields fused through a Cahn-Hilliard variational model. Vehicle control is formulated as a stochastic optimal control problem solved via forward-backward stochastic differential equations. The paper claims existence and uniqueness of a Nash equilibrium with exponential convergence (Theorems 1 and 2) and presents qualitative simulations for lane changing and overtaking scenarios with N=4 vehicles, asserting outperformance over optimization and game-theoretic baselines.

Significance. The problem of scalable, interaction-aware, safety-conscious trajectory planning is important, and the proposed combination of mean field games with physics-inspired potential fields is conceptually original. If the theoretical guarantees were established, the paper would make a meaningful contribution to the field. However, the central convergence proof is not substantiated: the contraction step in Theorem 1 is asserted without derivation, several model objects are left undefined, and the simulation section does not include the promised baselines. As written, the paper does not support its main claims of guaranteed Nash equilibrium, exponential convergence, or superior empirical performance.

major comments (3)
  1. [§3.6, Theorem 1 and Theorem 2] The proof's Step 4 asserts 'we can choose the parameters such that sqrt(L_Gamma) < 1' without deriving L_Gamma or exhibiting a parameter choice. Step 3 only states an inequality E[∫||u1*-u2*||²dt] ≤ L_Gamma W2²(m1,m2) and then concludes W2(Γ(m1),Γ(m2)) ≤ sqrt(L_Gamma) W2(m1,m2); this does not establish contractivity. Γ(m) requires solving the FBSDE system (22)-(25) for each m, with running cost (21) that includes −Φ(π(x)), where Φ* is itself defined through the m-dependent variational problems (11)-(19). No Lipschitz bound for the composed map is supplied, and no admissible parameter tuple satisfying sqrt(L_Gamma) < 1 is given. Theorem 2 reuses the same unsupported factor as ρ. Since existence, uniqueness, and exponential convergence are the paper's central theoretical claims, this gap is load-bearing and not a presentation issue.
  2. [§4, Simulation Results] The abstract and conclusion claim IUPF 'outperforms traditional optimization and game-theoretic baselines in both adaptability and computational efficiency,' but §4 reports only qualitative plots of IUPF trajectories for lane changing and overtaking. No baseline algorithms are named, no quantitative metrics (safety distance margins, computation time, success rate, smoothness/efficiency indices) are compared, and no statistical results over multiple runs are reported. The convergence rate ρ predicted by Theorem 2 is not measured. Additionally, the field parameters in §4.1.3 are 'calibrated through systematic tuning to achieve optimal performance' and then evaluated on the same scenarios, so the reported safety/efficiency is not an independent test of generalizability. Without a controlled comparison, the empirical contribution is limited to illustrative demonstrations.
  3. [§3.2–§3.5] Theorem 1's assumptions are not connected to the actual model. The running cost (21) contains ℓ_θ(x,m) and the terminal cost Ψ^i, neither of which is defined; the interaction kernel K_θ in Eq. (9) is never specified; and the style space Θ in Eq. (6) is not mapped to the style-dependent field parameters in Eq. (10). Moreover, Φ* is defined through the Cahn-Hilliard steady-state problem (17)-(19) with B*,R* depending on m, so Assumption 3 ('Φ* is bounded and twice continuously differentiable') and Assumption 1 (Lipschitz in m) require verification that is absent. These undefined objects make it impossible to check the claimed Lipschitz and strong-convexity hypotheses on which the contraction argument relies.
minor comments (6)
  1. [Theorem 1 heading] Contains typos: 'Uniqeness' and 'Eqilibrium' should be 'Uniqueness' and 'Equilibrium'.
  2. [§4.1.3] The symbol T is used both for the simulation horizon (T=15 s) and for the 'system temperature parameter' (T=0.1). This is confusing; please rename one of them.
  3. [Figure 1] The figure caption uses full-width parentheses '（a)' and '（b)' instead of ASCII parentheses. Please fix the formatting.
  4. [§3.4, §4] The numerical method for solving the variational problems (11)-(16) and the modified Cahn-Hilliard PDE (17) is not described, even though the simulations depend on these solutions.
  5. [Acknowledgments] The acknowledgments section contains the template placeholder 'If you wish to include any acknowledgments...' and should be removed or filled.
  6. [References] Several references are incomplete (e.g., [8], [9] lack venue/page information). Please check the bibliography formatting.

Circularity Check

0 steps flagged

No circularity found: the IUPF derivation is self-contained; Theorem 1's unproved contraction choice and tuned simulation parameters are correctness/evaluation gaps, not circular reductions.

full rationale

I walked the claimed derivation chain. The potential fields B*, R*, and Φ* are defined by the variational problems (11)-(19) without reference to the Nash equilibrium, and the FBSDE optimality system (22)-(25) is the standard necessary condition for the cost (20)-(21). The equilibrium is then defined as a fixed point of the best-response map Γ in §3.6, so the construction is not definitionally self-referential. The self-citations [26] and [27] are background examples of MFG methods requiring external safety critics; they do not supply the contraction or the equilibrium result. Two flagged weaknesses are real but are not circularity. First, Theorem 1 Step 4 asserts: "Under the strong convexity and Lipschitz assumptions, we can choose the parameters such that sqrt(L_Γ) < 1, making Γ a contraction mapping." No expression for L_Γ in terms of model parameters is derived, and no parameter tuple is exhibited, so the contraction property—and hence Theorem 2's exponential convergence, which reuses the same ρ—is unproved. This is an omitted proof/unsupported existential, not a reduction of the conclusion to the assumptions. Second, §4.1.3 states the field parameters are "calibrated through systematic tuning," and §4 reports no actual optimization/game-theoretic baseline comparisons, so the safety, efficiency, and outperformance claims are partly fitted and unbenchmarked. These are evaluation limitations, not definitional circularity. Under the strict standard of circularity—an equation or result being equivalent to its own inputs by construction—no such step is present.

Axiom & Free-Parameter Ledger

9 free parameters · 4 axioms · 4 invented entities

The framework's central claims depend on a large set of hand-tuned or un-specified parameters (field decay lengths, coupling strengths, control costs, style-parameter mappings) and on ad hoc assumptions such as the contraction of the best-response map. The invented entities (Cahn-Hilliard fusion, temperature, coupling function, interaction kernel) have no independent evidence.

free parameters (9)
  • benefit field decay length lambda_B = 155.0 m
    Calibrated in Section 4.1.3 to achieve 'long-range nature of traffic flow opportunities'; central to the shape of the benefit field which drives the planner.
  • risk field decay length lambda_R = 8.0 m
    Calibrated in Section 4.1.3; determines localized collision risk influence.
  • coupling strength kappa = 3.3
    Mentioned in Section 4.1.3 as governing nonlinear interaction between benefit and risk, but kappa never appears in the model equations; an unexplained parameter.
  • nonlinearity exponent alpha = 2.8
    Also stated in Section 4.1.3 but not connected to alpha_i in equation (18); unexplained.
  • system temperature T = 0.1
    Stated in Section 4.1.3 to regulate field sensitivity, but no temperature appears in any equation; an invented parameter.
  • process noise intensity sigma_w = 0.05
    Set in Section 4.1.3; affects stochastic dynamics and error bounds.
  • coupling parameters gamma_1..gamma_5, alpha_1, alpha_2 = not given
    Used in equation (18) but no values or calibration procedure provided; the fusion term is defined only symbolically.
  • style-dependent field parameters alpha_B, alpha_R, lambda_B, lambda_R, sigma_B, sigma_R per style = not specified
    Equation (10) defines them per style but the paper never defines the map theta -> parameters nor gives values for conservative/aggressive/cooperative styles.
  • control cost matrix R_theta = not given
    R_theta = diag(r_s, r_d) in equation (21) but values are never specified.
axioms (4)
  • domain assumption Existence and uniqueness of FBSDE solutions for the individual control problem given the mean field measure
    Invoked in Theorem 1 Step 1 via 'standard results in stochastic optimal control theory'; not proven or cited specifically.
  • domain assumption Lipschitz and strong convexity assumptions on cost and dynamics (Assumptions 1-3 of Theorem 1)
    Stated in Theorem 1; they are strong and not verified for the specific cost with -Phi(pi(x)) and the unspecified ell_theta.
  • ad hoc to paper The best-response operator Gamma is a contraction with sqrt(L_Gamma) < 1
    Asserted in Theorem 1 Step 4 without derivation; the paper says 'we can choose the parameters such that' but no such choice is specified.
  • ad hoc to paper The style space Theta maps to field parameters in a way that is not defined
    Equation (10) defines alpha_theta etc., but the mapping is never given; the simulation uses only global lambda values.
invented entities (4)
  • Unified Potential Field (IUPF) derived from Cahn-Hilliard dynamics no independent evidence
    purpose: Represents a single scalar field combining benefit and risk for the planner's cost function
    The Cahn-Hilliard fusion and coupling function chi with sin/cos terms are introduced ad hoc; no physical or empirical justification, and no independent prediction.
  • System temperature T no independent evidence
    purpose: Regulates field sensitivity
    Appears only in simulation parameters, never defined in the model equations.
  • Coupling function chi no independent evidence
    purpose: Nonlinear interaction between benefit and risk fields in equation (18)
    Involves raw tunable parameters gamma_i and arbitrary sin/cos terms; no derivation.
  • Interaction kernel K_theta in the mean-field drift no independent evidence
    purpose: Models mean field interaction in equation (9)
    Defined only abstractly; never specified or used in simulations.

pith-pipeline@v1.3.0-alltime-deepseek · 11132 in / 14154 out tokens · 135534 ms · 2026-08-04T21:10:10.701482+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Mean Field Game-Based Interactive Trajectory Planning Using Physics-Inspired Unified Potential Fields." pith.science (2026). https://pith.science/paper/52IAK2UR

@misc{pith2026250908147,
  author       = {Pith},
  title        = {Pith review of: Mean Field Game-Based Interactive Trajectory Planning Using Physics-Inspired Unified Potential Fields},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/52IAK2UR}},
  note         = {Machine review of arXiv:2509.08147}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Interactive trajectory planning in autonomous driving must balance safety, efficiency, and scalability under heterogeneous driving behaviors. Existing methods often face high computational cost or rely on external safety critics. To address this, we propose an Interaction-Enriched Unified Potential Field (IUPF) framework that fuses style-dependent benefit and risk fields through a physics-inspired variational model, grounded in mean field game theory. The approach captures conservative, aggressive, and cooperative behaviors without additional safety modules, and employs stochastic differential equations to guarantee Nash equilibrium with exponential convergence. Simulations on lane changing and overtaking scenarios show that IUPF ensures safe distances, generates smooth and efficient trajectories, and outperforms traditional optimization and game-theoretic baselines in both adaptability and computational efficiency.

Figures

Figures reproduced from arXiv: 2509.08147 by Chunhong Yuan, Fujiang Yuan, Yanhong Peng, Zhen Tian.

Figure 1
Figure 1. Figure 1: (a) Interaction between the HV and surrounding [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed IUPF framework. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Unified potential field evolution during lane changing maneuver: (a) Spatio-temporal field dynamics showing vehicle [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Unified potential field dynamics during overtaking maneuver. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

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

27 extracted references · 22 canonical work pages · 2 internal anchors

  1. [1]

    Te Chen, Yingfeng Cai, et al. 2023. Trajectory and Velocity Planning Method of Emergency Rescue Vehicle Based on Segmented Three-Dimensional Quartic Bezier Curve.IEEE Trans. Intell. Transp. Syst.s24, 3 (2023), 3461–3475. https: //doi.org/10.1109/TITS.2022.3224785

    work page arXiv 2023

  2. [2]

    Rong Guo and Shibo Yu. 2022. A Graph-search based Trajectory Planning Approach with Longitudianl Priority. InProceeding of the CAA International Conference on Vehicular Control and Intelligence. 1–6. https://doi.org/10.1109/ CVCI56766.2022.9964675

    arXiv 2022

  3. [3]

    Yuqing Guo, Danya Yao, et al. 2022. Trajectory Planning for an Autonomous Vehicle in Spatially Constrained Environments.IEEE Trans. Intell. Transp. Syst.s 23, 10 (2022), 18326–18336. https://doi.org/10.1109/TITS.2022.3164548

    work page arXiv 2022

  4. [4]

    Chao Huang, Hailong Huang, et al. 2022. Human-Machine Cooperative Trajectory Planning and Tracking for Safe Automated Driving.IEEE Trans. Intell. Transp. Syst.s23, 8 (2022), 12050–12063. https://doi.org/10.1109/TITS.2021.3109596

    work page arXiv 2022

  5. [5]

    Seulbin Hwang, Kibeom Lee, et al. 2022. Autonomous Vehicle Cut-In Algorithm for Lane-Merging Scenarios via Policy-Based Reinforcement Learning Nested Within Finite-State Machine.IEEE Trans. Intell. Transp. Syst.s23, 10 (2022), 17594–17606. https://doi.org/10.1109/TITS.2022.3153848

    work page arXiv 2022

  6. [6]

    Stefan Kojchev, Robert Hult, et al. 2024. A Two-Stage MIQP-Based Optimization Approach for Coordinating Automated Electric Vehicles in Confined Sites.IEEE Trans. Intell. Transp. Syst.s25, 2 (2024), 2061–2075. https://doi.org/10.1109/TITS. 2023.3320168

    arXiv 2024

  7. [7]

    Taeheon Lee and Seibum B. Choi. 2023. Real-Time Optimization of Gear Shift Trajectories Using Quadratic Programming for Electric Vehicles With Dual Clutch Transmission.IEEE Trans. Veh. Technol.72, 7 (2023), 8647–8660. https://doi.org/ 10.1109/TVT.2023.3246179

    work page arXiv 2023

  8. [8]

    Qinghao Li et al. 2025. Adaptive Field Effect Planner for Safe Interactive Au- tonomous Driving on Curved Roads.arXiv preprint arXiv:2504.14747(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  9. [9]

    Qinghao Li et al . 2025. Efficient and Safe Planner for Automated Driving on Ramps Considering Unsatisfication.arXiv preprint arXiv:2504.15320(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  10. [10]

    Zhihao Lin et al. 2025. Multi-Agent Monte Carlo Tree Search for Safe Decision Making at Unsignalized Intersections. (2025)

    work page 2025

  11. [11]

    Zhihao Lin, Jianglin Lan, Christos Anagnostopoulos, Zhen Tian, and David Flynn. 2025. Safety-Critical Multi-Agent MCTS for Mixed Traffic Coordination at Unsignalized Intersections.IEEE Transactions on Intelligent Transportation Systems(2025)

    work page 2025

  12. [12]

    Zhihao Lin, Zhen Tian, Qi Zhang, Hanyang Zhuang, and Jianglin Lan. 2024. Enhanced visual slam for collision-free driving with lightweight autonomous cars.Sensors24, 19 (2024), 6258

    work page 2024

  13. [13]

    Zhihao Lin, Qi Zhang, Zhen Tian, Peizhuo Yu, and Jianglin Lan. 2024. DPL-SLAM: enhancing dynamic point-line SLAM through dense semantic methods.IEEE Sensors Journal24, 9 (2024), 14596–14607

    work page 2024

  14. [14]

    Zhihao Lin, Qi Zhang, Zhen Tian, Peizhuo Yu, Ziyang Ye, Hanyang Zhuang, and Jianglin Lan. 2025. Slam2: Simultaneous localization and multimode mapping for indoor dynamic environments.Pattern Recognition158 (2025), 111054

    work page 2025

  15. [15]

    Yonggang Liu, Bobo Zhou, et al. 2021. Dynamic lane-changing trajectory planning for autonomous vehicles based on discrete global trajectory.IEEE Trans. Intell. Transp. Syst.s23, 7 (2021), 8513–8527

    work page 2021

  16. [16]

    Tianqi Qie, Weida Wang, et al. 2023. An Improved Model Predictive Control-Based Trajectory Planning Method for Automated Driving Vehicles Under Uncertainty Environments.IEEE Trans. Intell. Transp. Syst.s24, 4 (2023), 3999–4015. https: //doi.org/10.1109/TITS.2022.3230680

    work page arXiv 2023

  17. [17]

    Patrick Scheffe, Matheus V. A. Pedrosa, et al. 2023. Receding Horizon Control Using Graph Search for Multi-Agent Trajectory Planning.IEEE Trans. Control Syst. Technol.31, 3 (2023), 1092–1105. https://doi.org/10.1109/TCST.2022.3214718

    arXiv 2023

  18. [18]

    Robbin van Hoek, Jeroen Ploeg, et al. 2021. Cooperative Driving of Automated Vehicles Using B-Splines for Trajectory Planning.IEEE Trans. Intell. Veh.6, 3 (2021), 594–604. https://doi.org/10.1109/TIV.2021.3072679

    work page arXiv 2021

  19. [19]

    Mingqiang Wang, Lei Zhang, et al. 2022. A hybrid trajectory planning strategy for intelligent vehicles in on-road dynamic scenarios.IEEE Trans. Veh. Technol. 72, 3 (2022), 2832–2847

    work page 2022

  20. [20]

    Songtao Xie, Junyan Hu, et al . 2022. Distributed Motion Planning for Safe Autonomous Vehicle Overtaking via Artificial Potential Field.IEEE Trans. Intell. Transp. Syst.s23, 11 (2022), 21531–21547. https://doi.org/10.1109/TITS.2022. 3189741

    doi:10.1109/tits.2022 2022

  21. [21]

    Qingyuan Yang, Yunfeng Ai, et al. 2023. Decoupled real-time trajectory planning for multiple autonomous mining trucks in unloading areas.IEEE Trans. Intell. Veh.(2023)

    work page 2023

  22. [22]

    Fujiang Yuan et al. 2025. Bio-inspired hybrid path planning for efficient and smooth robotic navigation: F. Yuan et al.International Journal of Intelligent Robotics and Applications(2025), 1–31

    work page 2025

  23. [23]

    Ziyu Zhang, Chunyan Wang, et al. 2023. Path-speed decoupling planning method based on risk cooperative game for intelligent vehicles.IEEE Transactions on Transportation Electrification(2023)

    work page 2023

  24. [24]

    Zhiqiang Zhang, Lei Zhang, et al. 2023. An Enabling Trajectory Planning Scheme for Lane Change Collision Avoidance on Highways.IEEE Trans. Intell. Veh.8, 1 (2023), 147–158. https://doi.org/10.1109/TIV.2021.3117840

    work page arXiv 2023

  25. [25]

    Cong Zhao, Yifan Zhu, et al. 2022. A Novel Direct Trajectory Planning Approach Based on Generative Adversarial Networks and Rapidly-Exploring Random Tree. IEEE Trans. Intell. Transp. Syst.s23, 10 (2022), 17910–17921. https://doi.org/10. 1109/TITS.2022.3164391

    work page arXiv 2022

  26. [26]

    Liancheng Zheng, Zhen Tian, Yangfan He, Shuo Liu, Ke Gong, Huilin Chen, and Zhihao Lin. 2025. Enhanced Mean Field Game for Interactive Decision-Making with Varied Stylish Multi-Vehicles.arXiv preprint arXiv:2509.00981(2025)

    Pith/arXiv arXiv 2025

  27. [27]

    Liancheng Zheng, Xuemei Wang, Feng Li, Zebing Mao, Zhen Tian, Yanhong Peng, Fujiang Yuan, and Chunhong Yuan. 2025. A Mean-Field-Game-Integrated MPC-QP Framework for Collision-Free Multi-Vehicle Control.Drones9, 5 (2025), 375

    work page 2025