Pith. sign in

REVIEW 4 major objections 4 minor 30 references

SGTP: Sampling-based Game-Theoretic Planning for Real-Time Multi-Vehicle Autonomous Racing

T0 review · 4 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A game-theoretic sampling planner wins 95.24% of simulated races in 0.095s per decision.

desk verdict Solid, clearly-specified planning framework with a useful benchmark, but the headline win rate rests on an unablated last-mover advantage and a misleading completion metric. read the letter →

arxiv 2607.25388 v1 pith:N6PTIXS7 submitted 2026-07-28 cs.RO

Zhouheng Li , Fangguo Zhao , Mattia Piccinini , Baha Zarrouki , Yuan Gao , Zitong Shan , Johannes Betz , Chen Lv
show 1 more author
Lei Xie
This is my paper · ORCID
classification cs.RO
keywords autonomousracingmulti-vehicleplanninggametheoryiterativebestresponsesampling-basedtrajectoryGPUaccelerationcollisionavoidancefeasibilityselection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that game-theoretic planning for multi-vehicle racing can be made real-time by replacing optimization-based best responses with GPU-parallel sampling, ranking candidates by a game-aware cost, and then explicitly filtering for track boundaries and collisions. The central claim is that this combination yields diverse competitive behaviors—defensive blocking, wheel-to-wheel contesting, opportunistic overtaking—while keeping the mean planning time at 0.095 seconds. In the paper’s simulated benchmark of seven tracks, SGTP reports a 95.24% win rate, a 99.35% task-completion ratio, and collision-free wins in 100% of winning trials. A sympathetic reader would care because existing interaction-aware planners are either too slow or rely on soft penalties that allow infeasible maneuvers.

What carries the argument

The central object is the game-aware cost function plus an explicit feasibility-selection gate, wrapped in a sampling-based iterative best-response loop. The cost ranks each candidate trajectory against predicted opponent trajectories by contest-state maintenance (staying within a longitudinal contest window), longitudinal advantage (progress gap), blocking alignment (matching the opponent’s tail lateral offset), and safety separation (penalizing minimum distance below a threshold). Feasibility selection then enforces a minimum track-boundary clearance and a minimum collision distance, and selects the cheapest feasible candidate; the selected control warm-starts the next iteration, and the e

What would settle it

Run SGTP on a physical 1:43-scale car or in a high-fidelity simulator with true swept-volume collision detection on the same seven tracks, and compare collision rate and win rate. If any wheel-to-wheel maneuver that appears collision-free at the 0.1s waypoints is actually a contact, the safety claim fails. Alternatively, test with opponent speeds scaled to 1.0 rather than 0.9 and from non-rearmost starting positions; if the win rate collapses, the headline number is an artifact of the setup.

Watch

Extended reading notes

Core claim

On its own terms, the paper’s discovery is that iterative best response can be carried by sampling rather than optimization. Each vehicle holds opponents’ previous trajectories fixed, samples 128 control perturbations on a GPU, rolls them through a kinematic bicycle model, and scores them with a pairwise game-aware cost made of four terms: contest maintenance, longitudinal advantage, blocking alignment, and safety separation. The lowest-cost candidate that also passes explicit track-boundary and collision-threshold checks is selected; if none passes, a least-violation fallback is used. The paper reports that removing the game-aware cost cuts the win rate from 95.24% to 50%, and removing the

Load-bearing premise

The whole result rests on the discrete-time simulator with a kinematic bicycle model and waypoint-aligned collision checks being a faithful stand-in for real close-quarters racing; if it is not, the 95% win rate and collision-free record will not transfer to the real world.

Editorial extensions

If this is right

  • Real-time game-theoretic racing is within reach: a mean 0.095s planning time means decisions can be made at roughly 10 Hz during close interaction.
  • Explicit feasibility selection, not soft penalties, appears to be what allows long-duration collision-free racing; the ablation without it terminates in about 5 seconds.
  • The same planner can serve all vehicles simultaneously and scales to 10 agents with stable runtime in the reported trials.
  • The game-aware cost can be plugged into standard path-integral sampling frameworks, but the paper shows that such a variant is not safe unless the explicit feasibility gate is retained.
  • The released open benchmark gives a common ground for comparing multi-agent racing planners.

Reading between the lines

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

  • Because the collision check is applied to time-aligned waypoints rather than continuous swept volumes, increasing speed or lowering the waypoint rate could produce collisions that the reported simulation misses; a high-fidelity or physical test would settle whether the 100% collision-free win rate transfers.
  • The benchmark places the ego rearmost with opponents slowed to 0.9 of reference speed, so the 95.24% win rate likely overstates performance in symmetric or faster-opponent races; the paper’s own perturbed-prediction tests show interaction duration drops sharply under noise.
  • The four hand-tuned game-aware weights are a significant empirical component, and the paper does not report how sensitive the headline numbers are to their precise values; an ablation over these weights would clarify how much the result depends on manual tuning.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes SGTP, a sampling-based game-theoretic planner for multi-vehicle autonomous racing. It combines GPU-parallel sampling of control sequences and dynamics rollouts with an iterative best-response (IBR) loop, a game-aware trajectory cost that encodes contest, longitudinal advantage, blocking, and safety terms, and an explicit feasibility selection enforcing track-boundary and collision-avoidance constraints. The planning loop is followed by an additional best-response update for the ego vehicle. The authors report simulation results in F1TENTH Gym on seven tracks with three vehicles and 42 trials, claiming a 95.24% win rate, a 99.35% 'task-completion ratio', a mean computation time of 0.095 s, and favorable comparisons against eight baselines and two ablations, plus a scalability study up to 10 agents. The code and benchmark are released.

Significance. If the claims hold, SGTP is a useful engineering contribution: it replaces the nonconvex MPC subproblems of standard IBR with GPU-parallel sampling, and it couples a game-aware cost with hard feasibility selection. The open-source code and benchmark are concrete assets, and the 0.095 s mean computation time with up to 10 agents is genuinely encouraging. The ablations show plausible roles for the game-aware cost and feasibility selection. However, the central quantitative claims rest on a small, single-seed evaluation and on an asymmetric ego-only final response, and the headline 'completion ratio' is not what it appears to be. These issues need to be resolved before the paper supports its strongest statements.

major comments (4)
  1. [Algorithm 1, line 14; Section II-B; Table I] The ego-only final best-response step is a potential last-mover advantage. In Algorithm 1, after the IBR loop, only the ego performs an additional response against the opponents' final predictions; opponents do not receive a corresponding last move. Since the main evaluation uses the same planner for ego and opponents (Section IV-E), this asymmetry may materially inflate the 95.24% win rate. The paper does not state whether IBR-MPC or EVO-MPCC baselines were given the same extra step. The GA-IBR-MPPI ablation does not isolate this effect. Please report SGTP without line 14, or give the same final-response privilege to all planners, and discuss the difference.
  2. [Table I, footnote c; Abstract] The abstract's '99.35% task-completion ratio' is actually D_ratio = D_mean/(maximum trial time), not the fraction of trials that ran to completion. Since trials terminate on collision, a trial ending at 49.67 s on average is not equivalent to a 99.35% completion rate. This terminology materially overstates the headline result. Please report the actual fraction of trials reaching the maximum duration and rename D_ratio accordingly.
  3. [Section IV-E, Table I] The headline win-rate comparison is statistically fragile. All numbers come from 42 trials with no repeated seeds, confidence intervals, or significance tests. In particular, SGTP's 95.24% (40/42) versus EVO-MPCC's 92.86% (39/42) is a difference of one trial, well within the about 3.3% standard error of p≈0.95 at n=42. The ablations (50.00%, 57.14%) are also based on the same 42-trial sample. The claim that SGTP is 'best overall' needs either more trials, repeated-seed reporting, or a statistical test.
  4. [Section IV-B, Section IV-E] The evaluation protocol gives the ego an inherent advantage: the ego is always placed rearmost while opponent reference speeds are scaled by 0.9. This likely creates more overtaking opportunities than a symmetric or varied initial-condition protocol. It is acceptable as a stress test, but it should be reported as a limitation and supplemented with sensitivity analyses, e.g., varying the speed handicap or initial ordering, before claiming general superiority in 'highly interactive races.'
minor comments (4)
  1. [Section IV-C, Table I] CFW is defined as the collision-free rate 'among winning trials,' but the paper later says '100.00% wins without collisions.' This wording can be misread as an overall collision-free rate. Please clarify that non-winning trials may involve collisions or other outcomes.
  2. [Section IV-E] Reporting only mean, std, and max of computation time without the worst-case or a percentile can obscure real-time suitability. Since CT_max for SGTP is 0.102 s, close to the 10 Hz control period, a statement about the fraction of planning calls exceeding the period would be more informative.
  3. [Section II-B] The initialization in Algorithm 1, line 3, is described as 'distance-based obstacle avoidance' but no equation is supplied. Please specify the initialization cost or refer to a concrete method.
  4. [Section IV-B] There are a few typos and formatting issues: 'MapZooa' should be 'MapZoo'; some inline equations in the text are rendered as symbols (e.g., the contest window formula in the figure caption). Please proofread the final version.

Circularity Check

1 steps flagged · score 2.0 of 10

Qualitative behavior claims are encoded in the game-aware cost; the headline win-rate benchmark is empirical and not circular.

  1. self definitional [Section III-A (Eqs. 3-5), Section IV-D, Section V]
    "The key design insight is to progressively shape the interaction through complementary cost terms: the contest term preserves rollouts that remain within a meaningful racing window, the longitudinal term promotes forward advantage, the blocking term encourages defensive positioning... These results show that SGTP generates diverse competitive behaviors through IBR updates and game-aware rollout ranking, including blocking, wheel-to-wheel contesting, and opportunistic overtaking."

    The qualitative 'key finding' that the game-aware cost produces blocking, contesting, and overtaking is not an independent prediction: those behaviors are explicitly the optimization objective. Eq. (3) rewards terminal progress gaps inside scontest ('contest'), Eq. (4) rewards positive terminal progress advantage ('overtaking'), and Eq. (5) rewards lateral alignment with an opponent when leading ('blocking'). Observing these behaviors after optimizing a cost that explicitly encodes them is a consistency check, not a derived result. The headline win-rate comparison against external baselines is unaffected by this by-construction element.

full rationale

The paper is primarily an empirical benchmark study rather than a derivation. The headline 95.24% win rate is obtained from closed-loop simulation against eight externally published baselines plus ablations, so the central performance claim is measured, not fitted or self-referential. The one genuinely by-construction element is qualitative: the game-aware cost explicitly rewards contest maintenance (Eq. 3), longitudinal advantage (Eq. 4), and blocking alignment (Eq. 5), and the paper then presents blocking/contesting/overtaking as its 'key finding.' That is a self-definitional consistency check rather than a prediction. The EVO-MPCC baseline [28] is an overlapping-author SSRN preprint, but it is used as a benchmark, not as the justification of the central claim, so it is at most a minor self-citation and not load-bearing. The abstract's '99.35% task-completion ratio' is actually D_ratio = D_mean/50s (Table I footnote), a reporting concern rather than circularity. Overall, no central result reduces to its inputs by construction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The method's outputs are measurements on a tuned system, not a parameter-free derivation. The headline numbers depend on hand-set weights, thresholds, sample counts, and simulation-fidelity assumptions; none are derived or given uncertainty intervals.

free parameters (6)
  • Game-aware cost weights (w_contest, w_long, w_block, w_safety) = [1.0, 2.0, 10.0, 50.0]
    Manually tuned (Sec. IV-B); they directly define what counts as contesting/blocking and shape every headline behavior metric.
  • w_game = 60
    Manual balance between game-aware and tracking cost; changes selected trajectory.
  • Sampling noise (sigma_a, sigma_delta) = [0.335, 0.025]
    Hand-set; determines the rollout cloud coverage and therefore the quality of approximate best responses.
  • Thresholds (s_contest, s_role, mu_tail, g_safe, m_bd, d_col) = [8.0 m, 1.0 m, 0.3, 1.0 m, 0.515 m, 0.9 m]
    Hand-chosen; define interaction windows and feasibility margins; no sensitivity analysis.
  • Tracking weights (Q, R, S) = Q=[60,60,47.75,39.48], R=[8.43,20.0], S=[1.0,19.26]
    Fixed in Sec. IV-B; affects nominal trajectory tracking.
  • Sample count/horizon/iterations (K, H, L_IBR) = [128, 12, 2]
    Design choices; larger K/H or more iterations would change trade-off between quality and 10 Hz runtime.
assumptions (5)
  • domain assumption Kinematic bicycle model f(·) [25] is accurate for planning and closed-loop simulation.
    Used in all rollouts and in the F1TENTH simulator; if inaccurate at racing limits, safety/performance claims weaken.
  • domain assumption Minimum-curvature raceline G [26] is a suitable reference for all vehicles and tracks.
    All vehicles track this reference; a poor raceline would disadvantage all planners uniformly but shape the scenarios.
  • domain assumption F1TENTH Gym [9] faithfully reproduces multi-vehicle racing interactions.
    All quantitative claims are generated in this simulator; no real-vehicle validation is provided.
  • domain assumption Fixed-opponent IBR with L_IBR=2 approximates a unilateral best response (Algorithm 1 lines 4-13).
    No convergence proof is given; two iterations may not reach an equilibrium and the 'game-theoretic' claim rests on this approximation.
  • domain assumption Discrete time-aligned distance checks with threshold d_col=0.9 m guarantee collision avoidance (Eq. 6, 11-12).
    Checks occur at H sampled steps, not continuous time; inter-sample collisions are not excluded.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SGTP: Sampling-based Game-Theoretic Planning for Real-Time Multi-Vehicle Autonomous Racing." pith.science (2026). https://pith.science/paper/N6PTIXS7

@misc{pith2026260725388,
  author       = {Pith},
  title        = {Pith review of: SGTP: Sampling-based Game-Theoretic Planning for Real-Time Multi-Vehicle Autonomous Racing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N6PTIXS7}},
  note         = {Machine review of arXiv:2607.25388}
}
read the original abstract

Autonomous multi-vehicle racing requires real-time planning of diverse competitive behaviors in intense interactions. Existing planners often struggle to balance strategic diversity and computational efficiency. To address this challenge, we propose Sampling-based Game-Theoretic Planning (SGTP), a real-time framework that combines game-theoretic reasoning with GPU-accelerated sampling of control sequences and dynamics rollouts. Sampled trajectories are ranked using a game-aware cost to capture competitive interactions and generate diverse racing behaviors. Our planner then performs feasibility selection by explicitly enforcing track-boundary and dynamic collision-avoidance constraints, ensuring safe and reliable transitions between racing strategies. Extensive simulations on challenging tracks show that SGTP achieves a 95.24% win rate and a 99.35% task-completion ratio in highly interactive races, with a mean computational time of 0.095 s over multiple iterative solving steps. We also demonstrate the successful application of SGTP in large-scale scenarios with up to 10 agents. We release our code and provide an open-source benchmark of multi-agent autonomous racing algorithms to facilitate future research. Project page: https://sgtp-racing.github.io/.

Figures

Figures reproduced from arXiv: 2607.25388 by the authors.

Figure 1
Figure 1. Proposed SGTP framework for multi-behavior autonomous racing. SGTP combines GPU-accelerated sampling-based planning with game-theoretic best-response reasoning, using a new game-aware cost to favor competitive interactions and selecting the lowest-cost feasible trajectory. interaction-aware reasoning to explore diverse strategies while ensuring reliable transitions between them. Game theory provides a principled fra… view at source ↗
Figure 2
Figure 2. Proposed Sampling-Based Game-Theoretic Planning (SGTP) framework. SGTP first generates sampled rollouts and ranks them using a game-aware cost evaluated against opponent predictions. During IBR, each vehicle computes a best response, after which feasibility selection enforces the track-boundary and inter-vehicle collision-avoidance constraints. The ego vehicle then performs an additional best-response update. Here, … view at source ↗
Figure 3
Figure 3. SGTP enables competitive behaviors and reliable transitions between them in highly interactive scenarios on multiple racetracks. The ego vehicle demonstrates diverse racing behaviors, including trailing and wheel-to-wheel contesting, before completing an opportunistic overtake after sustained close-range interactions. Videos are available on the project webpage. E. Quantitative Results: Performance & Real-Time Effic… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Longitudinal gaps and velocity profiles from [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Competitive performance & scalability of SGTP. Left plot: our SGTP achieves the best balance across the normalized metrics of Sec. IV-C. Right plot: SGTP maintains low, consistent computation times and collision-free operation as the number of vehicles increases, outpe…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 4 linked inside Pith

  1. [1]

    Autonomous vehicles on the edge: A survey on autonomous vehicle racing,

    J. Betz, H. Zheng, A. Liniger, U. Rosolia, P. Karle, M. Behl, V . Krovi, and R. Mangharam, “Autonomous vehicles on the edge: A survey on autonomous vehicle racing,”IEEE Open Journal of Intelligent Trans- portation Systems, vol. 3, pp. 458–488, 2022

    2022

  2. [2]

    Forzaeth race stack—scaled autonomous head-to-head racing on fully commercial off-the-shelf hardware,

    N. Baumann, E. Ghignone, J. K ¨uhne, N. Bastuck, J. Becker, N. Imholz, T. Kr¨anzlin, T. Y . Lim, M. L¨otscher, L. Schwarzenbachet al., “Forzaeth race stack—scaled autonomous head-to-head racing on fully commercial off-the-shelf hardware,”Journal of Field Robotics, vol. 42, no. 4, pp. 1037–1079, 2025

    2025

  3. [3]

    Modular decision-making and drivable areas for multi-agent autonomous racing,

    A. Toschi, F. Prignoli, and M. Bertogna, “Modular decision-making and drivable areas for multi-agent autonomous racing,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2025, pp. 12 435–12 441

    2025

  4. [4]

    Optimization-based au- tonomous racing of 1: 43 scale rc cars,

    A. Liniger, A. Domahidi, and M. Morari, “Optimization-based au- tonomous racing of 1: 43 scale rc cars,”Optimal Control Applications and Methods, vol. 36, no. 5, pp. 628–647, 2015

    2015

  5. [5]

    Reduce lap time for autonomous racing with curvature-integrated mpcc local trajectory planning method,

    Z. Li, L. Xie, C. Hu, and H. Su, “Reduce lap time for autonomous racing with curvature-integrated mpcc local trajectory planning method,” in 2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), 2024, pp. 1066–1073

    2024

  6. [6]

    A data-driven aggressive autonomous racing framework utilizing local trajectory planning with velocity prediction,

    Z. Li, B. Zhou, C. Hu, L. Xie, and H. Su, “A data-driven aggressive autonomous racing framework utilizing local trajectory planning with velocity prediction,” in2025 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2025, pp. 16 657–16 663

    2025

  7. [7]

    Kineto-dynamical planning and accurate execution of minimum-time maneuvers on three-dimensional circuits,

    M. Piccinini, S. Taddei, J. Betz, and F. Biral, “Kineto-dynamical planning and accurate execution of minimum-time maneuvers on three-dimensional circuits,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 1–7

    2025

  8. [8]

    A multi-stage time-variant motion planner for agile autonomous driving maneuvers,

    A. Langmann, S. Kohl, L. ¨Ogretmen, M. Piccinini, and J. Betz, “A multi-stage time-variant motion planner for agile autonomous driving maneuvers,”IEEE Open Journal of Intelligent Transportation Systems, vol. 7, pp. 1136–1150, 2026

    2026

Show all 30 references
  1. [9]

    End2race: Efficient end- to-end imitation learning for real-time f1tenth racing,

    Z. Qiao, H. Li, Z. Cao, and H. X. Liu, “End2race: Efficient end- to-end imitation learning for real-time f1tenth racing,”arXiv preprint arXiv:2509.16894, 2025

    2025

  2. [10]

    Flow matching-based autonomous driving planning with advanced interactive behavior modeling,

    T. Tan, Y . Zheng, R. Liang, Z. Wang, K. Zheng, J. Zheng, J. Li, X. Zhan, and J. Liu, “Flow matching-based autonomous driving planning with advanced interactive behavior modeling,” inAdvances in Neural Information Processing Systems, vol. 38. Curran Associates, Inc., 2025, pp....

    2025

  3. [11]

    Diffusion-based planning for autonomous driving with flexible guidance,

    Y . Zheng, R. Liang, K. Zheng, J. Zheng, L. Mao, J. Li, W. Gu, R. Ai, S. Li, X. Zhanet al., “Diffusion-based planning for autonomous driving with flexible guidance,” inInternational Conference on Learning Representations, vol. 2025, 2025, pp. 37 207–37 227

    2025

  4. [12]

    Rapid and safe trajectory planning over diverse scenes through diffusion composition,

    W. Mao, Z. Li, Y . Luo, Y . Du, and L. Xie, “Rapid and safe trajectory planning over diverse scenes through diffusion composition,”arXiv preprint arXiv:2507.04384, 2025

    2025 arXiv

  5. [13]

    Strategizing at speed: A learned model predictive game for multi-agent drone racing,

    A.-C. Papuc, L. Peters, S. Sun, L. Ferranti, and J. Alonso-Mora, “Strategizing at speed: A learned model predictive game for multi-agent drone racing,”arXiv preprint arXiv:2602.06925, 2026

    2026 arXiv

  6. [14]

    α-racer: Real-time algorithm for game-theoretic motion planning and control in autonomous racing using near-potential function,

    D. Kalaria, C. Maheshwari, and S. Sastry, “α-racer: Real-time algorithm for game-theoretic motion planning and control in autonomous racing using near-potential function,” inConference on Learning for Dynamics & Control, 2024

    2024

  7. [15]

    Learning two-agent motion planning strategies from generalized nash equilibrium for model predictive control,

    H. Kim, E. L. Zhu, C. S. Lim, and F. Borrelli, “Learning two-agent motion planning strategies from generalized nash equilibrium for model predictive control,” inProceedings of the 7th Annual Learning for Dynamics & Control Conference, ser. Proceedings of Machine Learning R...

    2025

  8. [16]

    Driving is a game: Combining planning and prediction with bayesian iterative best response,

    A. Distelzweig, Y . Wang, F. Janjo ˇs, M. Hallgarten, M. Dobre, A. Lang- mann, J. Boedecker, and J. Betz, “Driving is a game: Combining planning and prediction with bayesian iterative best response,” inPro- ceedings of the 19th European Conference on Computer Vision (ECCV 2026...

    2026

  9. [17]

    Regulation-aware game-theoretic motion planning for autonomous racing,

    F. Prignoli, F. Borrelli, P. Falcone, and M. Pustilnik, “Regulation-aware game-theoretic motion planning for autonomous racing,”arXiv preprint arXiv:2508.20203, 2025

    2025 arXiv

  10. [18]

    A sequential quadratic programming approach to the solution of open-loop generalized nash equilibria,

    E. L. Zhu and F. Borrelli, “A sequential quadratic programming approach to the solution of open-loop generalized nash equilibria,” in2023 IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 3211–3217

    2023

  11. [19]

    A real-time game theoretic planner for autonomous two-player drone racing,

    R. Spica, E. Cristofalo, Z. Wang, E. Montijano, and M. Schwager, “A real-time game theoretic planner for autonomous two-player drone racing,”IEEE Transactions on Robotics, vol. 36, no. 5, pp. 1389–1403, 2020

    2020

  12. [20]

    Game- theoretic planning for self-driving cars in multivehicle competitive scenarios,

    M. Wang, Z. Wang, J. Talbot, J. C. Gerdes, and M. Schwager, “Game- theoretic planning for self-driving cars in multivehicle competitive scenarios,”IEEE Transactions on Robotics, vol. 37, no. 4, pp. 1313– 1325, 2021

    2021

  13. [21]

    A rapid iterative trajectory planning method for automated parking through differential flatness,

    Z. Li, L. Xie, C. Hu, and H. Su, “A rapid iterative trajectory planning method for automated parking through differential flatness,”Robotics and Autonomous Systems, vol. 182, p. 104816, 2024

    2024

  14. [22]

    Model predictive control via probabilistic inference: A tutorial,

    K. Honda, “Model predictive control via probabilistic inference: A tutorial,”arXiv preprint arXiv:2511.08019, 2025

    2025 arXiv

  15. [23]

    Biased-mppi: Informing sampling- based model predictive control by fusing ancillary controllers,

    E. Trevisan and J. Alonso-Mora, “Biased-mppi: Informing sampling- based model predictive control by fusing ancillary controllers,”IEEE Robotics and Automation Letters, vol. 9, no. 6, pp. 5871–5878, 2024

    2024

  16. [24]

    Stein variational guided model predictive path integral control: Proposal and experiments with fast maneuvering vehicles,

    K. Honda, N. Akai, K. Suzuki, M. Aoki, H. Hosogaya, H. Okuda, and T. Suzuki, “Stein variational guided model predictive path integral control: Proposal and experiments with fast maneuvering vehicles,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE...

    2024

  17. [25]

    The kinematic bicycle model: A consistent model for planning feasible trajectories for autonomous vehicles?

    P. Polack, F. Altch ´e, B. d’Andr ´ea Novel, and A. de La Fortelle, “The kinematic bicycle model: A consistent model for planning feasible trajectories for autonomous vehicles?” in2017 IEEE Intelligent Vehicles Symposium (IV), 2017, pp. 812–818

    2017

  18. [26]

    Minimum curvature trajectory planning and control for an autonomous race car,

    A. Heilmeier, A. Wischnewski, L. Hermansdorfer, J. Betz, M. Lienkamp, and B. Lohmann, “Minimum curvature trajectory planning and control for an autonomous race car,”Vehicle System Dynamics, 2020

    2020

  19. [27]

    Motion plan- ning for autonomous driving with a conformal spatiotemporal lattice,

    M. McNaughton, C. Urmson, J. M. Dolan, and J.-W. Lee, “Motion plan- ning for autonomous driving with a conformal spatiotemporal lattice,” in2011 IEEE International Conference on Robotics and Automation, 2011, pp. 4889–4895

    2011

  20. [28]

    Evo-mpcc: Enhanced velocity optimization with learning-based auto-tuning for real-time vehicle trajectory planning,

    Z. Li, B. Zhou, M. Piccinini, C. Hu, B. Zarrouki, R. Mangharam, and L. Xie, “Evo-mpcc: Enhanced velocity optimization with learning-based auto-tuning for real-time vehicle trajectory planning,”Available at SSRN 6127037, 2026

    2026

  21. [29]

    CasADi – A software framework for nonlinear optimization and opti- mal control,

    J. A. E. Andersson, J. Gillis, G. Horn, J. B. Rawlings, and M. Diehl, “CasADi – A software framework for nonlinear optimization and opti- mal control,”Mathematical Programming Computation, vol. 11, no. 1, pp. 1–36, 2019

    2019

  22. [30]

    On adversarial robustness of trajectory prediction for autonomous vehicles,

    Q. Zhang, S. Hu, J. Sun, Q. A. Chen, and Z. M. Mao, “On adversarial robustness of trajectory prediction for autonomous vehicles,” inPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 15 159–15 168

    2022

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.