Pith. sign in

REVIEW 3 major objections 4 minor 3 cited by

TrajEvo: Designing Trajectory Prediction Heuristics via LLM-driven Evolution

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read An LLM inside an evolutionary loop writes trajectory-prediction heuristics that beat handcrafted rules in-distribution and transfer to unseen data better than neural models.

desk verdict TrajEvo is a credible, well-scoped application of LLM-driven evolution to trajectory prediction, but its headline claim of beating deep learning on SDD rests on an under-documented Table 3 and single stochastic runs. read the letter →

arxiv 2505.04480 v1 pith:WB74RZGW submitted 2025-05-07 cs.AI cs.NEcs.RO

Zhikai Zhao , Chuanbo Hua , Federico Berto , Kanghoon Lee , Zihan Ma , Jiachen Li , Jinkyoo Park This is my paper · ORCID
classification cs.AIcs.NEcs.RO
keywords trajectorypredictionlargelanguagemodelsevolutionaryalgorithmsautomatedheuristicdesigncross-datasetgeneralizationCross-GenerationEliteSamplingStatisticsFeedbackLoopmulti-agentpedestrian
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

TrajEvo tries to establish that trajectory-prediction heuristics—short, readable pieces of code—can be designed automatically by a large language model inside an evolutionary loop, instead of being handcrafted by experts or encoded in large neural networks. The framework seeds a population with a constant-velocity predictor, then repeatedly evaluates, reflects on, recombines, and mutates candidate Python heuristics on the ETH-UCY benchmark. If its results hold, the payoff is practical: evolved heuristics match or beat handcrafted heuristics in-distribution, and on the unseen Stanford Drone Dataset they outperform recent neural models in average-position error while running about 300 times faster on a single CPU core. The authors present this as a first step toward automatically designed, explainable, and generalizable prediction rules.

What carries the argument

The engine is a reflective evolutionary loop, adapted from the Reflective Evolution paradigm, in which an LLM implements the genetic operators: it writes the initial population from a constant-velocity seed function, performs crossover by mixing code from two parents under textual reflections, and mutates an elite candidate. Two mechanisms carry the novel argument. Cross-Generation Elite Sampling is a mutation-target selector: instead of always mutating the current best heuristic, it keeps a history of high-performing heuristics and samples the one to mutate with a Softmax over their recorded objective values, a design meant to escape local optima. The Statistics Feedback Loop is a per-slot diagnostic: after evaluating $K=20$ predicted trajectories per agent, it counts how often each of the 20 slots produced the lowest average displacement on individual instances and feeds that distribution, together with the heuristic code, into the reflection and mutation prompts. That lets the LLM discover which diversification strategies actually earn their keep—for instance, the evolved Zara1 heuristic reserves one slot for a near-deterministic linear extrapolation. The objective being minimized is $J = 0.6\, \mathrm{minADE}_{20} + 0.4\, \mathrm{minFDE}_{20}$.

What would settle it

Independently train EigenTrajectory and MoFlow on the same ETH-UCY leave-one-out splits used for TrajEvo and evaluate them on SDD with identical frame scaling, units, $K=20$ sampling, and code infrastructure. If either model then matches or beats TrajEvo's average $\mathrm{minADE}_{20}/\mathrm{minFDE}_{20}$ of $12.65/24.14$ pixels, the paper's headline claim—that evolved heuristics outperform deep learning under distribution shift—is falsified.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that simple automatically evolved code can transfer across scene distributions better than neural networks trained on the source distribution. Evolved on ETH-UCY, TrajEvo's best heuristics report an average $\mathrm{minADE}_{20}/\mathrm{minFDE}_{20}$ of $0.36/0.71$ m on the ETH-UCY test sets, ahead of every heuristic baseline it compares against. In the headline cross-dataset experiment, applying heuristics evolved on each ETH-UCY split to the unseen SDD gives an average $\mathrm{minADE}_{20}/\mathrm{minFDE}_{20}$ of $12.65/24.14$ pixels, versus $14.55/25.71$ for EigenTrajectory and $17.14/28.56$ for MoFlow. The paper is explicit that in-distribution the strongest recent neural models still lead; the claimed advantage is specifically on unseen data, plus speed and explainability. Separately, the authors show that the two added mechanisms—Cross-Generation Elite Sampling and the Statistics Feedback Loop—each improve the search, and they report inference at 0.65 ms per instance on a single CPU core.

Load-bearing premise

The headline cross-dataset comparison assumes the deep learning baselines were trained and evaluated under exactly the same leave-one-out splits, preprocessing, units, and $K=20$ protocol as TrajEvo, a protocol Section 4.3 does not describe.

Editorial extensions

If this is right

  • On ETH-UCY, the best evolved heuristic averages $\mathrm{minADE}_{20}/\mathrm{minFDE}_{20}$ of $0.36/0.71$ m, placing it ahead of every heuristic baseline listed in Table 1 on every dataset.
  • Under leave-one-out training on ETH-UCY and testing on the unseen SDD, evolved heuristics average $\mathrm{minADE}_{20}/\mathrm{minFDE}_{20}$ of $12.65/24.14$ px, better than EigenTrajectory ($14.55/25.71$) and MoFlow ($17.14/28.56$).
  • Generated heuristics run at 0.65 ms per instance on one CPU core, more than 300 times faster than the CPU inference time reported for MoFlow, making real-time CPU-only deployment plausible.
  • Ablation of either the Statistics Feedback Loop or Cross-Generation Elite Sampling raises ETH-UCY errors, so both mechanisms contribute to the evolved heuristics' quality.
  • A full evolution run costs roughly $0.05 in API usage and about five minutes, two orders of magnitude cheaper than the reported one-day GPU training of neural baselines.

Reading between the lines

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

  • The same evolution recipe could be applied to other forecasting tasks where kinematic baselines still compete, such as vehicle motion or sports tracking; the paper only evaluates pedestrian datasets.
  • The Statistics Feedback Loop's observed benefit of keeping one deterministic linear extrapolation among the 20 samples suggests a general design principle for multimodal predictors: reserve a 'safe' anchor slot before sampling diversity. This is an interpretation, not a claim the paper tests directly.
  • Because the evolved heuristics are plain Python, they could be compiled or translated for embedded deployment; the paper notes C++ conversion as future work and reports a preliminary more than 20x speedup from a zero-shot translation request.
  • If the cross-dataset result holds under a shared protocol, it would indicate that simple inductive biases written as code can be more transferable than learned representations—a causal claim the paper's experiments do not directly establish.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes TrajEvo, a framework that uses an LLM (Gemini 2.0 Flash) as the generator and reflector inside an evolutionary loop to produce Python-coded trajectory prediction heuristics. The main algorithmic additions are Cross-Generation Elite Sampling (CGES), which samples mutation targets from a history archive, and a Statistics Feedback Loop, which feeds per-sample prediction statistics to the reflector. The heuristics are evaluated on the ETH-UCY datasets under a leave-one-out protocol and on the Stanford Drone Dataset (SDD) for cross-dataset generalization. The central claims are that TrajEvo outperforms existing heuristic baselines on ETH-UCY, outperforms deep learning baselines on the unseen SDD dataset, and does so with substantially lower training and inference cost while remaining interpretable.

Significance. If the reported results are reproducible, TrajEvo is a valuable demonstration that LLM-driven evolutionary search can discover non-neural predictors that are fast, interpretable, and competitive with or better than deep baselines under distribution shift. The paper ships executable heuristic code and a public repository, and the evaluation loop is concrete enough to be machine-checked. The cross-dataset SDD experiment is the most interesting and potentially important result. However, the headline claim currently rests on an under-specified baseline comparison and on single stochastic runs, so the significance is conditional on the authors providing a complete protocol and variance-aware results. The work is not circular: evolution optimizes on training splits and evaluation is on held-out splits.

major comments (3)
  1. [§3.1] The multi-sample metric is described in a non-standard and ambiguous way. The text says the method selects a single set index k* that minimizes an unspecified “overall error” and then uses the trajectories of that set for all agents, rather than the usual per-pedestrian best-of-K protocol used in the cited trajectory prediction literature. If the deep learning baseline numbers in Tables 2 and 3 were obtained with per-agent minADE20/minFDE20 while TrajEvo uses a global set index, all comparisons are not apples-to-apples. Please define “overall error,” state explicitly which protocol is used for every method in every table, and, if the two protocols differ, report both sets of values.
  2. [§4.3, Table 3] The cross-dataset generalization claim—the central contribution stated in the abstract—is supported only by Table 3, but the table gives no experimental protocol for the deep learning baselines. There is no statement of whether Trajectron++, EigenTrajectory, and MoFlow were trained by the authors on the same ETH-UCY leave-one-out splits, no checkpoint provenance, no description of SDD preprocessing (frame-rate resampling, observation/prediction horizons, units, agent filtering, coordinate conventions), and no details on how K=20 sampling was performed for each baseline. Without these details, the “outperforms deep learning on unseen SDD” claim cannot be independently verified. Please provide a complete protocol and, ideally, run all baselines in the same code infrastructure.
  3. [§4.2–§4.4, Table 5] The evolutionary search is stochastic: the LLM temperature is 1, CGES uses softmax sampling with temperature 1, and crossover and mutation are randomized, yet every reported result is a single run with no seeds or error bars. The margins in Tables 1–4, including the roughly 1.9-pixel minADE advantage over EigenTrajectory on SDD, could be within run-to-run variation. Please re-run each configuration with multiple seeds and report the mean, standard deviation (or full distribution), and the number of seeds used.
minor comments (4)
  1. [§4.5, Appendix A.2] The inference-time comparison that supports the 300× speedup claim lacks measurement details for the neural baselines: no batch size, no number of CPU cores used for the multi-core CPU timings, no GPU warm-up protocol, and no statement of whether the same preprocessing and code paths were used. Please specify these details.
  2. [Appendix B.2, Prompt 12] The prompts inform the LLM that the task uses the ETH/UCY dataset, but the paper does not address the possibility that Gemini's pretraining has already seen SDD trajectories. Since SDD is described as “unseen,” the authors should discuss this leakage risk and, if possible, test it by comparing against an LLM that has no exposure to such data or by analyzing whether the evolved heuristics encode SDD-specific patterns.
  3. [Figure 3] The right panel of Figure 3 plots objective value over function evaluations for “TrajEvo w/o CGES” and “TrajEvo,” but the curves have no error bars, no axis labels with units, and no indication of how many runs were averaged; the claimed benefit of CGES is therefore not statistically substantiated.
  4. [Table 3] The repeated header “A VG→SDD” is confusing, and it is not explained why the heuristic baselines (SocialForce, CVM, CVM-S) are constant across the ETH, HOTEL, UNIV, ZARA1, and ZARA2 columns while the neural baselines vary; please clarify the split-dependence of each row.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the evolutionary objective is a training metric, test and SDD evaluations are held out, and the ReEvo self-citation is methodological rather than load-bearing.

full rationale

The derivation is self-contained. TrajEvo evolves Python heuristics by evaluating them on training splits with the standard minADE20/minFDE20 objective, then evaluates the final heuristics on held-out ETH-UCY test splits and on the unseen SDD dataset. Because the evolutionary fitness is computed on training data, using the same metric for test evaluation is ordinary supervised model selection, not a reduction of the prediction to its inputs. The Statistics Feedback Loop similarly feeds training-split statistics back into the LLM; no test labels are used. The only self-citation is to ReEvo [38] as the base evolutionary framework; this citation is methodological rather than load-bearing for the claimed result, whose evidence is the reported cross-dataset experiments. Concerns about whether the Table 3 deep-learning baselines were run under an identical protocol are evidence and reproducibility issues, not definitional circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper is an empirical engineering contribution rather than a derivation, so the ledger focuses on search-level hand-set parameters and on unstated evaluation assumptions. The final heuristics' internal constants are chosen by the LLM during evolution, not by the authors, so they are not free parameters in the classic sense. The main transfer-risk assumptions are fair baseline evaluation and absence of LLM pretraining leakage.

free parameters (2)
  • Objective weights wADE and wFDE = 0.6 and 0.4
    Chosen by hand in Section 3.1 to define the combined fitness J = 0.6 * minADE20 + 0.4 * minFDE20; the evolutionary search optimizes this J, so the chosen weights shape which heuristics are selected.
  • Evolution hyperparameters (population size, initial generation, elite ratio, crossover rate, mutation rate, CGES… = population 10, initial generation 8, elite ratio 0.3, crossover rate 1, mutation rate 0.5, CGES softmax temp 1.0, LLM…
    Hand-set in Table 5; the paper says some were 'empirically tuned for the trajectory prediction task' but does not report validation-based selection, so the results may vary with these settings.
assumptions (4)
  • domain assumption The deep learning baselines in Table 3 were trained and evaluated under the same protocol as TrajEvo
    Section 4.3 gives no protocol details, so this premise is load-bearing and unverified; if false, the cross-dataset superiority claim collapses.
  • domain assumption Gemini 2.0 Flash reliably generates executable, syntactically valid code and coherent reflections
    The entire evolutionary loop depends on LLM code generation; no alternative LLM is tested, so the result may be LLM-specific.
  • ad hoc to paper The LLM's pretraining knowledge does not leak SDD-specific motion patterns into generated heuristics
    Prompt 12 names ETH/UCY as the dataset, but the pretrained LLM may have seen SDD or similar drone footage, which would weaken the 'unseen' claim; the paper does not control for this.
  • domain assumption The multi-sample minADE20/minFDE20 protocol is an appropriate and fair evaluation for both heuristics and neural baselines
    Standard protocol from Social-GAN, but it rewards methods that produce one good trajectory out of 20; the comparison assumes all methods are evaluated under equivalent sampling conditions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TrajEvo: Designing Trajectory Prediction Heuristics via LLM-driven Evolution." pith.science (2026). https://pith.science/paper/WB74RZGW

@misc{pith2026250504480,
  author       = {Pith},
  title        = {Pith review of: TrajEvo: Designing Trajectory Prediction Heuristics via LLM-driven Evolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WB74RZGW}},
  note         = {Machine review of arXiv:2505.04480}
}
read the original abstract

Trajectory prediction is a crucial task in modeling human behavior, especially in fields as social robotics and autonomous vehicle navigation. Traditional heuristics based on handcrafted rules often lack accuracy, while recently proposed deep learning approaches suffer from computational cost, lack of explainability, and generalization issues that limit their practical adoption. In this paper, we introduce TrajEvo, a framework that leverages Large Language Models (LLMs) to automatically design trajectory prediction heuristics. TrajEvo employs an evolutionary algorithm to generate and refine prediction heuristics from past trajectory data. We introduce a Cross-Generation Elite Sampling to promote population diversity and a Statistics Feedback Loop allowing the LLM to analyze alternative predictions. Our evaluations show TrajEvo outperforms previous heuristic methods on the ETH-UCY datasets, and remarkably outperforms both heuristics and deep learning methods when generalizing to the unseen SDD dataset. TrajEvo represents a first step toward automated design of fast, explainable, and generalizable trajectory prediction heuristics. We make our source code publicly available to foster future research at https://github.com/ai4co/trajevo.

Figures

Figures reproduced from arXiv: 2505.04480 by the authors.

Figure 1
Figure 1. Motivation for TRAJEVO. Traditional manual heuristic design (left) is based on human experts with trial and error. Deep learning (center) generates better predictions but requires significant computational resources, generates black-box models, and struggles with generalization. TRAJEVO (right) automates the design process of heuristics via evolutionary algorithms, generating novel trajectory prediction heuristics. … view at source ↗
Figure 2
Figure 2. Example evolution of trajectory prediction heuristics with T [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Cross-Generation Elite Sampling (CGES) helps escape local optima by sampling elite individuals [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Statistics obtained by running TRAJEVO-generated code are provided alongside the corresponding code as an input to the reflector (left). TRAJEVO then analyses this and gathers insights into how to evolve better trajectory prediction heuristics (right). This loop specif…
Figure 5
Figure 5. Figure 5: Comparison of trajectory prediction results between CVM-S [ [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. USPR: Learning a Unified Solver for Profiled Routing

    cs.LG 2025-05 conditional novelty 6.0 of 10

    A unified transformer-based reinforcement-learning policy, USPR, encodes arbitrary vehicle-client profile scores and profile weights and outperforms prior neural PVRP solvers on synthetic, large-scale, and CVRPLib-der...

  2. Large Foundation Models for Trajectory Prediction in Autonomous Driving: A Comprehensive Survey

    cs.RO 2025-09 conditional novelty 4.0 of 10

    A structured survey of LLM-based trajectory prediction methods, organized into trajectory-language mapping, multimodal fusion, and constraint-based reasoning, with benchmarks, metrics, and future directions.

  3. Large Language Models for Next-Generation Wireless Network Management: A Survey and Tutorial

    cs.NI 2025-09 conditional novelty 4.0 of 10

    A survey and tutorial that organizes LLM-enabled wireless network optimization into formulation, solution, and verification stages, with case studies drawn from the authors' own prior papers.

Reference graph

Works this paper leans on

74 extracted references · 57 canonical work pages · cited by 3 Pith papers

  1. [1]

    N. A. Madjid, A. Ahmad, M. Mebrahtu, Y . Babaa, A. Nasser, S. Malik, B. Hassan, N. Werghi, J. Dias, and M. Khonji. Trajectory prediction for autonomous driving: Progress, limitations, and future directions. arXiv preprint arXiv:2503.03262, 2025

    arXiv 2025

  2. [2]

    S. Wang, Z. Chen, Z. Zhao, C. Mao, Y . Zhou, J. He, and A. S. Hu. Escirl: Evolving self- contrastive irl for trajectory prediction in autonomous driving. In 8th Annual Conference on Robot Learning, 2024

    work page 2024

  3. [3]

    Robla-G ´omez, V

    S. Robla-G ´omez, V . M. Becerra, J. R. Llata, E. Gonz ´alez-Sarabia, C. Torre-Ferrero, and J. P ´erez-Oria. Working together: A review on safe human-robot collaboration in industrial environments. IEEE Access, 5:26754–26773, 2017. doi:10.1109/ACCESS.2017.2773127

    arXiv 2017

  4. [4]

    L. P. Vishwakarma, R. K. Singh, R. Mishra, D. Demirkol, and T. Daim. The adoption of social robots in service operations: a comprehensive review.Technology in Society, 76:102441, 2024

    work page 2024

  5. [5]

    P. Li, X. Pei, Z. Chen, X. Zhou, and J. Xu. Human-like motion planning of autonomous vehicle based on probabilistic trajectory prediction. Applied Soft Computing, 118:108499, 2022

    work page 2022

  6. [6]

    Nakamura, R

    K. Nakamura, R. Tian, and A. Bajcsy. Not all errors are made equal: A regret metric for detect- ing system-level trajectory prediction failures. In 8th Annual Conference on Robot Learning, 2024. 9

    work page 2024

  7. [7]

    Mavrogiannis, F

    C. Mavrogiannis, F. Baldini, A. Wang, D. Zhao, P. Trautman, A. Steinfeld, and J. Oh. Core challenges of social robot navigation: A survey. ACM Transactions on Human-Robot Interac- tion, 12(3):1–39, 2023

    2023

  8. [8]

    Mahdi, S

    H. Mahdi, S. A. Akgun, S. Saleh, and K. Dautenhahn. A survey on the design and evolution of social robots—past, present and future. Robotics and Autonomous Systems , 156:104193, 2022

    work page 2022

Show all 74 references
  1. [9]

    Amirian, B

    J. Amirian, B. Zhang, F. V . Castro, J. J. Baldelomar, J.-B. Hayet, and J. Pettr ´e. Opentraj: Assessing prediction complexity in human trajectories datasets. In Proceedings of the asian conference on computer vision, 2020

    2020

  2. [10]

    Helbing and P

    D. Helbing and P. Molnar. Social force model for pedestrian dynamics. Physical review E, 51 (5):4282, 1995

    1995

  3. [11]

    Luber, J

    M. Luber, J. A. Stork, G. D. Tipaldi, and K. O. Arras. People tracking with human motion predictions from social forces. In 2010 IEEE International Conference on Robotics and Au- tomation, pages 464–469, 2010. doi:10.1109/ROBOT.2010.5509779

    2010

  4. [12]

    Zanlungo, T

    F. Zanlungo, T. Ikeda, and T. Kanda. Social force model with explicit collision prediction. Europhysics Letters, 93(6):68005, 2011

    2011

  5. [13]

    Farina, D

    F. Farina, D. Fontanelli, A. Garulli, A. Giannitrapani, and D. Prattichizzo. Walking ahead: The headed social force model. PloS one, 12(1):e0169734, 2017

    2017

  6. [14]

    X. Chen, M. Treiber, V . Kanagaraj, and H. Li. Social force models for pedestrian traffic–state of the art. Transport reviews, 38(5):625–653, 2018

    2018

  7. [15]

    van den Berg, M

    J. van den Berg, M. Lin, and D. Manocha. Reciprocal velocity obstacles for real-time multi- agent navigation. In 2008 IEEE International Conference on Robotics and Automation, pages 1928–1935, 2008. doi:10.1109/ROBOT.2008.4543489

    2008

  8. [16]

    Y . Ma, D. Manocha, and W. Wang. Efficient reciprocal collision avoidance between heteroge- neous agents using ctmat. In AAMAS, 2018

    2018

  9. [17]

    Yamaguchi, A

    K. Yamaguchi, A. C. Berg, L. E. Ortiz, and T. L. Berg. Who are you with and where are you going? In CVPR 2011, pages 1345–1352, 2011. doi:10.1109/CVPR.2011.5995468

    2011

  10. [18]

    Alahi, K

    A. Alahi, K. Goel, V . Ramanathan, A. Robicquet, L. Fei-Fei, and S. Savarese. Social lstm: Human trajectory prediction in crowded spaces. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 961–971, 2016

    2016

  11. [19]

    Salzmann, B

    T. Salzmann, B. Ivanovic, P. Chakravarty, and M. Pavone. Trajectron++: Dynamically-feasible trajectory forecasting with heterogeneous data. In Computer Vision–ECCV 2020: 16th Eu- ropean Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XVIII 16 , pages 683–700. Sp...

    2020

  12. [20]

    J. Li, C. Hua, J. Park, H. Ma, V . Dax, and M. J. Kochenderfer. Evolvehypergraph: Group-aware dynamic relational reasoning for trajectory prediction.arXiv preprint arXiv:2208.05470, 2022

    2022 arXiv

  13. [21]

    B. A. Rainbow, Q. Men, and H. P. Shum. Semantics-stgcnn: A semantics-guided spatial- temporal graph convolutional network for multi-class trajectory prediction. In 2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC), pages 2959–2966. IEEE, 2021

    2021

  14. [22]

    Gupta, J

    A. Gupta, J. Johnson, L. Fei-Fei, S. Savarese, and A. Alahi. Social gan: Socially acceptable trajectories with generative adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2255–2264, 2018. 10

    2018

  15. [23]

    Dendorfer, S

    P. Dendorfer, S. Elflein, and L. Leal-Taix´e. Mg-gan: A multi-generator model preventing out- of-distribution samples in pedestrian trajectory prediction. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13158–13167, 2021

    2021

  16. [24]

    S. Kim, J. Baek, J. Kim, and J. Lee. Guide-cot: Goal-driven and user-informed dynamic estimation for pedestrian trajectory using chain-of-thought. arXiv preprint arXiv:2503.06832, 2025

    2025 arXiv

  17. [25]

    I. Bae, J. Lee, and H.-G. Jeon. Can language beat numerical regression? language-based multimodal trajectory prediction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 753–766, 2024

    2024

  18. [26]

    Y . Yang, P. Zhu, M. Qi, and H. Ma. Uncovering the human motion pattern: Pattern memory- based diffusion model for trajectory prediction. arXiv preprint arXiv:2401.02916, 2024

    2024 arXiv

  19. [27]

    T. Gu, G. Chen, J. Li, C. Lin, Y . Rao, J. Zhou, and J. Lu. Stochastic trajectory prediction via motion indeterminacy diffusion. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17113–17122, 2022

    2022

  20. [28]

    Y . Fu, Q. Yan, L. Wang, K. Li, and R. Liao. Moflow: One-step flow matching for human trajectory forecasting via implicit maximum likelihood estimation based distillation. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),

  21. [29]

    Itkina and M

    M. Itkina and M. Kochenderfer. Interpretable self-aware neural networks for robust trajectory prediction. In Conference on Robot Learning, pages 606–617. PMLR, 2023

    2023

  22. [30]

    Jiang, K

    J. Jiang, K. Yan, X. Xia, and B. Yang. A survey of deep learning-based pedestrian trajectory prediction: Challenges and solutions. Sensors (Basel, Switzerland), 25(3):957, 2025

    2025

  23. [31]

    P. Liu, H. Liu, Y . Li, T. Shi, M. Zhu, and Z. Pu. Traj-explainer: An explainable and robust multi-modal trajectory prediction approach. arXiv preprint arXiv:2410.16795, 2024

    2024 arXiv

  24. [32]

    Cai and Z

    Y . Cai and Z. Ren. Pwto: A heuristic approach for trajectory optimization in complex terrains. arXiv preprint arXiv:2407.02745, 2024

    2024 arXiv

  25. [33]

    Korbmacher and A

    R. Korbmacher and A. Tordeux. Review of pedestrian trajectory prediction methods: Compar- ing deep learning and knowledge-based approaches. IEEE Transactions on Intelligent Trans- portation Systems, 23(12):24126–24144, 2022

    2022

  26. [34]

    Rudenko, L

    A. Rudenko, L. Palmieri, M. Herman, K. M. Kitani, D. M. Gavrila, and K. O. Arras. Human motion trajectory prediction: A survey. The International Journal of Robotics Research , 39 (8):895–935, 2020

    2020

  27. [35]

    Phong, H

    T. Phong, H. Wu, C. Yu, P. Cai, S. Zheng, and D. Hsu. What truly matters in trajectory prediction for autonomous driving? Advances in Neural Information Processing Systems, 36: 71327–71339, 2023

    2023

  28. [36]

    F. Liu, X. Tong, M. Yuan, and Q. Zhang. Algorithm evolution using large language model. arXiv preprint arXiv:2311.15249, 2023

    2023 arXiv

  29. [37]

    P. V . T. Dat, L. Doan, and H. T. T. Binh. Hsevo: Elevating automatic heuristic design with diversity-driven harmony search and genetic algorithm using llms. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 26931–26938, 2025

    2025

  30. [38]

    H. Ye, J. Wang, Z. Cao, F. Berto, C. Hua, H. Kim, J. Park, and G. Song. Reevo: Large lan- guage models as hyper-heuristics with reflective evolution. InAdvances in Neural Information Processing Systems, 2024. 11

    2024

  31. [39]

    Z. Chen, Z. Zhou, Y . Lu, R. Xu, L. Pan, and Z. Lan. Uber: Uncertainty-based evolution with large language models for automatic heuristic design. arXiv preprint arXiv:2412.20694, 2024

    2024 arXiv

  32. [40]

    F. Liu, X. Tong, M. Yuan, X. Lin, F. Luo, Z. Wang, Z. Lu, and Q. Zhang. Evolution of heuristics: Towards efficient automatic algorithm design using large language model. arXiv preprint arXiv:2401.02051, 2024

    2024 arXiv

  33. [41]

    Yuksekgonul, F

    M. Yuksekgonul, F. Bianchi, J. Boen, S. Liu, P. Lu, Z. Huang, C. Guestrin, and J. Zou. Optimiz- ing generative ai by backpropagating language model feedback. Nature, 639(8055):609–616, 2025

    2025

  34. [42]

    F. Liu, R. Zhang, Z. Xie, R. Sun, K. Li, X. Lin, Z. Wang, Z. Lu, and Q. Zhang. Llm4ad: A platform for algorithm design with large language model. arXiv preprint arXiv:2412.17287, 2024

    2024

  35. [43]

    F. Liu, Y . Yao, P. Guo, Z. Yang, Z. Zhao, X. Lin, X. Tong, M. Yuan, Z. Lu, Z. Wang, et al. A systematic survey on large language models for algorithm design. arXiv preprint arXiv:2410.14716, 2024

    2024

  36. [44]

    Wu, S.-h

    X. Wu, S.-h. Wu, J. Wu, L. Feng, and K. C. Tan. Evolutionary computation in the era of large language model: Survey and roadmap. IEEE Transactions on Evolutionary Computation , 2024

    2024

  37. [45]

    Zhang, F

    R. Zhang, F. Liu, X. Lin, Z. Wang, Z. Lu, and Q. Zhang. Understanding the importance of evolutionary search in automated heuristic design with large language models. InInternational Conference on Parallel Problem Solving from Nature, pages 185–202. Springer, 2024

    2024

  38. [46]

    Sch ¨oller, V

    C. Sch ¨oller, V . Aravantinos, F. Lay, and A. Knoll. What the constant velocity model can teach us about pedestrian motion prediction. IEEE Robotics and Automation Letters , 5(2):1696– 1703, 2020

    2020

  39. [47]

    Polychronopoulos, M

    A. Polychronopoulos, M. Tsogas, A. J. Amditis, and L. Andreone. Sensor fusion for predicting vehicles’ path for collision avoidance systems. IEEE Transactions on Intelligent Transporta- tion Systems, 8(3):549–562, 2007

    2007

  40. [48]

    T. Lu, Y . Watanabe, S. Yamada, and H. Takada. Comparative evaluation of kalman filters and motion models in vehicular state estimation and path prediction. Journal of Navigation, 74, 06

  41. [49]

    G. Zhai, H. Meng, and X. Wang. A constant speed changing rate and constant turn rate model for maneuvering target tracking. Sensors, 14(3):5239–5253, 2014. ISSN 1424-8220. doi: 10.3390/s140305239. URL https://www.mdpi.com/1424-8220/14/3/5239

    2014 doi

  42. [50]

    Helbing, I

    D. Helbing, I. Farkas, and T. Vicsek. Modeling the dynamics of human behavior in complex systems. Physical Review E, 56(4):4282, 1997

    1997

  43. [51]

    Huang, H

    Y . Huang, H. Bi, Z. Li, T. Mao, and Z. Wang. Stgat: Modeling spatial-temporal interactions for human trajectory prediction. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2019

    2019

  44. [52]

    Mohamed, K

    A. Mohamed, K. Qian, M. Elhoseiny, and C. Claudel. Social-stgcnn: A social spatio-temporal graph convolutional neural network for human trajectory prediction. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14424–14432, 2020

    2020

  45. [53]

    C. Xu, W. Mao, W. Zhang, and S. Chen. Remember intentions: Retrospective-memory-based trajectory prediction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6488–6497, 2022. 12

    2022

  46. [54]

    I. Bae, J. Oh, and H.-G. Jeon. Eigentrajectory: Low-rank descriptors for multi-modal trajectory forecasting. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 10017–10029, 2023

    2023

  47. [55]

    Liang, W

    J. Liang, W. Huang, F. Xia, P. Xu, K. Hausman, B. Ichter, P. Florence, and A. Zeng. Code as policies: Language model programs for embodied control. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 9493–9500. IEEE, 2023

    2023

  48. [56]

    B. Chen, R. Zhang, B. Wang, I. Zhang, K. Wang, Y . Chen, L. Xu, K. Tang, T. Zhou, S. Zhang, et al. Natural language is all you need for code generation. arXiv preprint arXiv:2310.02061, 2023

    2023 arXiv

  49. [57]

    C. Xia, T. Zhang, Z. Wang, H. Li, T. Jiang, T. Wang, H. Zhang, T. Li, S. Zhang, B. Wang, et al. Natural language to code: Can llms write better code than humans? arXiv preprint arXiv:2305.14750, 2023

    2023 arXiv

  50. [58]

    Romera-Paredes, M

    B. Romera-Paredes, M. Barekatain, A. Novikov, M. Balog, M. P. Kumar, E. Dupont, F. J. Ruiz, J. S. Ellenberg, P. Wang, O. Fawzi, et al. Mathematical discoveries from program search with large language models. Nature, 625(7995):468–475, 2024

    2024

  51. [59]

    Pryzant, D

    R. Pryzant, D. Iter, J. Li, Y . T. Lee, C. Zhu, and M. Zeng. Automatic prompt optimization with” gradient descent” and beam search. arXiv preprint arXiv:2305.03495, 2023

    2023 arXiv

  52. [60]

    E. C. Osuna and D. Sudholt. Runtime analysis of probabilistic crowding and restricted tour- nament selection for bimodal optimisation. In Proceedings of the Genetic and Evolutionary Computation Conference, pages 929–936, 2018

    2018

  53. [61]

    Pellegrini, A

    S. Pellegrini, A. Ess, K. Schindler, and L. van Gool. You’ll never walk alone: Modeling social behavior for multi-target tracking. In 2009 IEEE 12th International Conference on Computer Vision (ICCV), pages 261–268, 2009. doi:10.1109/ICCV .2009.5459260

    2009

  54. [62]

    Lerner, Y

    A. Lerner, Y . Chrysanthou, and D. Lischinski. Crowds by example. In Computer graphics forum, volume 26, pages 655–664. Wiley Online Library, 2007

    2007

  55. [63]

    C. M. Bishop and N. M. Nasrabadi. Pattern recognition and machine learning , volume 4. Springer, 2006

    2006

  56. [64]

    Gemini api pricing — gemini api — google ai for developers

    Google AI for Developers. Gemini api pricing — gemini api — google ai for developers. https://ai.google.dev/gemini-api/docs/pricing, apr 2025. Last updated: 2025-04- 21, Accessed: 2025-05-01

    2025

  57. [65]

    ```python ... ```

    A. Robicquet, A. Sadeghian, A. Alahi, and S. Savarese. Learning social etiquette: Human trajectory understanding in crowded scenes. In European Conference on Computer Vision (ECCV), pages 549–565. Springer International Publishing, 2016. 13 Supplementary Materials A Experiment...

    2016

  58. [68]

    Then, apply agent-specific stochastic variations within those constraints

    **Hierarchical Stochasticity:** Sample trajectory-level parameters (speed scale, movement pattern) *once* per trajectory. Then, apply agent-specific stochastic variations within those constraints. Introduce `global_randomness` sampled *once* per trajectory to couple different ...

  59. [69]

    Consider longer history windows

    **Adaptive Movement Primitives:** Condition movement model probabilities (stop, turn, straight, lane change, obstacle avoidance) on agent state (speed, acceleration, past turning behavior, context). Consider longer history windows

  60. [70]

    Experiment with learnable parameters and wider ranges

    **Refine Noise & Parameters:** Finetune noise scales and apply dampening. Experiment with learnable parameters and wider ranges. Directly manipulate velocity and acceleration stochastically for smoother transitions. 19

  61. [71]

    **Contextual Interactions:** Enhance social force models, considering intentions, agent types, and environment

  62. [72]

    **Guaranteed Diversity:** Ensure movement probabilities sum to 1

  63. [73]

    **Post Processing:** Apply smoothing and collision avoidance

  64. [74]

    going to an area

    **Intentions:** Incorporate high level intentions such as "going to an area." Output 14: Long-term reasoning output Output 14 shows an example of long-term reasoning output for the model, based on the comparative analysis. T RAJ EVO discovers several interesting heuristics for...

  65. [2021]

    doi:10.1017/S0373463321000370

    doi

  66. [2025]

    URL https://arxiv.org/abs/2503.09950

    arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.