Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

DriveAgent: Multi-Agent Structured Reasoning with LLM and Multimodal Sensor Fusion for Autonomous Driving

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read DriveAgent claims that a modular hierarchy of LLM-driven agents—one per sensor modality, coordinated by a final response agent—achieves better vehicle and environmental reasoning than monolithic prompting baselines on a new three-route…

desk verdict The dataset and the cross-sensor discrepancy-checking idea are worth engaging with; the performance numbers are not yet trustworthy until the fine-tuning/reasoning split is clarified. read the letter →

arxiv 2505.02123 v1 pith:F3THU3Q5 submitted 2025-05-04 cs.RO cs.DB

Xinmeng Hou , Wuqi Wang , Long Yang , Hao Lin , Jinglun Feng , Haigen Min , Xiangmo Zhao This is my paper · ORCID
classification cs.ROcs.DB
keywords multi-agentreasoningautonomousdrivingmultimodalsensorfusionvision-languagemodelslargelanguageLiDAR-camerasituationalunderstandingtimestampfiltration
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 proposes DriveAgent, a multi-agent autonomous driving framework in which a large language model coordinates specialized agents that reason over camera, LiDAR, GPS, and IMU data. It claims that structuring reasoning into four sequential modules—selection of critical timestamps, vehicle-level diagnostics from vision and LiDAR, environmental and causal analysis, and urgency-aware response generation—improves situational understanding. On the authors' new three-route dataset, DriveAgent beats prompting baselines: object detection F1 of 71.62 percent versus 68.80 for the best baseline, vision-based vehicle reasoning accuracy of 96.84 percent on one route, and environmental reasoning of 65.71 percent. A reader should care because the design targets interpretable, sensor-robust driving reasoning rather than end-to-end planning alone.

What carries the argument

The load-bearing mechanism is the Module 1 self-referential timestamp filter, which classifies a route by speed and urban complexity and selects critical timestamps where angular velocity, linear acceleration, or yaw rate exceed hand-set kinematic baselines (10 degrees per second, 8 meters per second squared, 10 degrees per second). All later vehicle and environmental reasoning operates only on these selected timestamps, so this filter determines the input distribution for the entire downstream pipeline. The vehicle reasoning module then relies on per-object position comparisons between consecutive frames and on the Euclidean distance between LiDAR and camera positions for each object, while environmental reasoning uses temporal differences in detection sets and a longer-interval state change to separate self-moving from externally influenced objects.

What would settle it

On the released dataset, enumerate every manually annotated critical event (e.g., a pedestrian on the road, a hard brake, or the onset of a camera misalignment) and check whether each appears among the timestamps selected by Module 1's kinematic filter; if any such event is missing, the downstream vehicle and environmental reasoning accuracies are computed over a filtered subset, and the claim of superior situational understanding is falsified.

Watch

Extended reading notes

Core claim

DriveAgent's central claim is that a modular hierarchy of LLM-driven agents, each consuming a single sensor modality, can perform vehicle and environmental reasoning more accurately than monolithic prompting strategies. The pipeline first filters sensor streams into a small number of critical timestamps, then has a vision agent and a LiDAR agent independently describe objects and motions, an aggregator diagnoses discrepancies, an environmental agent detects changes between timestamps, and a causal agent explains them. Using a fine-tuned LLaMA-3.2-Vision model as the vision-language backbone, the system reports the highest F1 on object and category detection, the highest accuracy on visual vehicle reasoning (including on deliberately misaligned left and right camera views), and the highest accuracy on environmental reasoning among the methods compared.

Load-bearing premise

The whole pipeline assumes the initial timestamp filter, which uses hand-set thresholds for turning speed, acceleration, and yaw rate, catches every moment that matters; if it misses an important event, no later agent gets a chance to reason about it.

Editorial extensions

If this is right

  • If the reported gains are real, the four-module pipeline gives a reusable recipe: a filter selects moments, modality-specific agents describe them, an aggregator diagnoses conflicts, and a final agent ranks urgent issues.
  • The fine-tuned vision-language model and structured annotation guidelines raise object-detection F1 from 45.55 for the base LLaMA-3.2-Vision model to 71.62, meaning annotation quality and instruction tuning carry much of the detection improvement.
  • Because the system outputs an explicit priority issue, a chosen action, and remaining insights, the design directly supports interpretable and auditable driving decisions rather than only final control commands.
  • The environment-reasoning results indicate that combining temporal difference detection with causal attribution beats chain-of-thought and self-refinement prompting on this dataset.

Reading between the lines

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

  • A next experiment the paper does not run is to measure Module 1's timestamp filter with precision and recall against manually labeled critical events; if the filter is lossy, the reasoning accuracies in Table IV may overstate situational understanding.
  • The cross-sensor consistency metric suggests a controlled test: artificially rotate or shift the camera images and verify that reasoned sensor-misalignment diagnoses and vision-reasoning accuracy respond systematically; the side-camera results hint at this but are not a controlled ablation.
  • The same agent-hierarchy pattern could transfer to other sensor-rich, safety-critical domains such as warehouse robotics or drone inspection, since the paper's contribution is a structured-reasoning template rather than a domain-specific planner.
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

4 major / 5 minor

Summary. DriveAgent proposes a modular, multi-agent LLM/VLM framework for autonomous driving situational understanding. The pipeline consists of four modules: descriptive event filtering (Module 1), vehicle-level reasoning from LiDAR and vision agents with an aggregator (Module 2), environmental and causal reasoning (Module 3), and urgency-aware response generation (Module 4). The authors introduce a new three-route dataset with synchronized camera, LiDAR, GPS, and IMU data, fine-tune LLaMA-3.2-Vision with LoRA, and evaluate object/category detection on a held-out route (R1) as well as vehicle and environmental reasoning on R2/R3. The main reported results are an F1 improvement on object detection (71.62 vs. 68.80 best baseline, Table III) and high reasoning accuracies on the new benchmarks (e.g., 96.84% vision reasoning on R2, 65.71% environmental reasoning on R3, Table IV).

Significance. If the claims are validated, DriveAgent would be a useful step toward interpretable, sensor-fusion-based reasoning for autonomous driving, combining structured modular decomposition with fine-tuned VLM perception. The paper's concrete assets are the new three-tier dataset, the released code and dataset links, and the held-out object detection comparison against strong commercial/LLM baselines, where the reported gain is a genuine contribution. However, the central claim of 'superior performance' from the multi-agent architecture is not yet established: the reasoning evaluations lack a stated train/test split and error analysis, and the baselines do not isolate the effect of the proposed architecture from the effect of VLM fine-tuning. These issues are fixable but require additional experiments and transparency, so the manuscript is not ready in its present form.

major comments (4)
  1. [III-B, III-D; Table IV] No train/test separation is reported for the vehicle and environmental reasoning tasks. The object detection task is explicitly stated to be trained on R2/R3 and evaluated on R1, but the reasoning tasks are described as 'evaluated by comparing the model's output with ground-truth labels in R2' and 'assessed on R2 and R3', and the environmental reasoning task states no split at all. Since the VLM is fine-tuned on instruction data collected from the same routes (Section III-D), the reasoning accuracies in Table IV, including the headline 96.84% on R2, could reflect memorization of the evaluation moments rather than the proposed multi-agent reasoning. The baselines are zero-shot prompting methods, so the comparison conflates fine-tuning gains with the framework's contribution. The authors must state a clear split (e.g., train on R2/R3, evaluate on R1, or a temporal split within each route) and include a fine-tuned single-agent VLM baseline that uses the same data without the multi-agent pipeline.
  2. [IV-B, Table IV] The claim that DriveAgent 'surpasses baseline prompting approaches in both accuracy and stability' is not supported by the full table. In the LiDAR 'Vehicle Understanding' rows, CoT + Self-Refine outperforms DriveAgent on R2 (72.63 vs. 65.26), R3 (64.76 vs. 55.24), and R2-right (63.89 vs. 58.33); in several vision rows, baseline methods also beat DriveAgent. No error bars, confidence intervals, significance tests, or repeated runs are reported, so differences such as 58.95 vs. 56.84 in environmental reasoning may be noise. The paper should report variance or significance and should temper the 'superior performance' claim to cases where the advantage actually holds.
  3. [II-A, Eqs. (1)-(2)] Module 1's timestamp filter is the sole gate for all downstream reasoning, but its accuracy is never evaluated. The thresholds θ_i are tailored from hand-set kinematic baselines (10°/s angular velocity, 8 m/s² linear acceleration, 10°/s yaw rate) with no precision/recall analysis and no ablation over threshold values. Because Modules 2-4 only receive timestamps that pass this filter, a low-recall filter would silently drop critical events and invalidate the reported vehicle and environment reasoning accuracies. The authors should evaluate the filter against a human-annotated set of critical events and report recall/precision, and ablate the threshold choices.
  4. [IV-A, Table III; III-B] The object detection experiment also does not isolate the proposed multi-agent architecture: it compares a fine-tuned DriveAgent VLM against zero-shot frozen VLMs. While this is a valid demonstration that fine-tuning with structured annotations helps, it does not support the abstract's attribution of gains to the multi-agent structured reasoning pipeline. An additional experiment with the same fine-tuned VLM used directly (without the multi-agent modules, or without the LiDAR/vision integration) is needed to identify which component contributes the improvement.
minor comments (5)
  1. [IV-B and III-B] There are several typographical errors: 'We first We first' and 'introduced in in Section III-B' should be fixed, and the conclusion's 'can further contributed' is ungrammatical.
  2. [Table IV caption] The caption refers to 'R-Left*' and 'R-Right*', but the table columns are labeled 'R2-left', 'R2-right', etc., and no explanation of the asterisk is given in the body or the caption.
  3. [References] Several references appear to be placeholders: [5] cites 'arXiv preprint arXiv:2401.12345', [8] cites 'arXiv preprint arXiv:2502.12345', and [13] and [14] cite the same paper title under different numbers. These need to be replaced with verifiable citations.
  4. [II-B, Eq. (4)] Equation (4) writes Δi(t) = Li(t) ∼ Ci(t), but Algorithm 1 uses the same symbol '∼' to denote position change between frames; this notation is confusing. Use an explicit norm, e.g., ∥Li(t) − Ci(t)∥.
  5. [III-C and II-D] The model name 'Pixtra-large' is likely 'Pixtral-Large'; also, the candidate set Φ and the functions Score and Ψ in Module 4 are used without definitions, so the reader cannot reproduce the response-generation step.

Circularity Check

1 steps flagged · score 6.0 of 10

Reasoning evaluation is confounded by VLM fine-tuning on the same routes; the R1 object-detection task remains a partially independent anchor.

  1. fitted input called prediction [Section III-B (Task and Evaluation Metrics), Section III-D (VLM Implementation Details), and Table IV (Reasoning Performance)]
    "The vehicle-reasoning task include two tasks: a LiDAR understanding task, evaluated by comparing the model’s output with ground-truth labels in R2, and a vision-based reasoning task, assessed on R2 and R3, where misaligned camera views serve as distractors. ... Training is performed for 10 epoch using instruction-style supervision introduced in the VLM Instructions section, where each training sample is formatted as an instruction-response pair that includes special<Image>tokens to denote visual inputs."

    The VLM backbone of DriveAgent is fine-tuned with instruction-style supervision on image-text prompt pairs (Section III-D), and the vehicle and environmental reasoning accuracies in Table IV are measured on the same routes R2 and R3 that the paper uses for the detection training split. The paper states a train/evaluation split only for object detection ('trained on datasets R2 and R3 and evaluated on R1'); no analogous split is given for the reasoning tasks. Consequently, the reported 'predictions' for vehicle reasoning (e.g., 96.84% on R2) and environmental reasoning are evaluated on inputs drawn from the model's fine-tuning distribution.

full rationale

The strongest independent result is the object-detection evaluation on R1, which is explicitly held out from the R2/R3 training split and therefore provides a real, non-circular anchor for the fine-tuned VLM's perception ability. However, the central reasoning claims in Table IV are subject to route-level training/evaluation overlap because the paper does not state any train/test split for the vehicle or environmental reasoning tasks, while the VLM is trained on multimodal instruction pairs whose content includes the same type of images used on R2 and R3. That overlap means the reasoning accuracies may be measuring memorization or in-distribution fitting rather than the multi-agent reasoning architecture. No other step in the pipeline—the kinematic timestamp filter, the LiDAR/camera consistency measures, or the utility-maximizing response generation—reduces to its own inputs by construction; those are ordinary algorithmic components. The self-citation to the prior data-collection work [17] is not load-bearing. Overall, the paper has genuine independent content, but the reasoning evaluation is partially circular because the fitted VLM is tested on its own training routes without a stated split, which warrants a score of 6 rather than a clean 0-2 result.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central empirical claim rests on hand-set kinematic thresholds and the 100 m LiDAR range cutoff as free parameters, plus assumptions that the upstream perception and the author-generated ground-truth labels are reliable. There are no new physical entities. The evaluation is self-contained, so the main risks are parameter choice and label validity rather than invented constructs.

free parameters (3)
  • Kinematic event thresholds in Module 1 = angular velocity 10°/s, linear acceleration 8 m/s², yaw rate 10°/s
    Eq. (2) sets standard kinematic baselines used by G(S,U,r_i) to compute θ_i; these are hand-set constants, not derived from data, and the paper provides no sensitivity analysis for them.
  • LiDAR range threshold R = 100 m
    Algorithm 1 and Section II-B3 restrict analysis to objects with ∥Li(t)∥≤100 m; this cutoff is chosen ad hoc and affects which object comparisons are used for vehicle diagnostics.
  • Causal analysis interval Δt = unspecified
    Eq. (6) uses an interval Δt for change detection but its value is never given; without it the environmental reasoning results are not reproducible.
assumptions (3)
  • ad hoc to paper The timestamp filter in Module 1 correctly identifies all critical events via R(S,U) and θ_i=G(S,U,r_i).
    Section II-A defines three route categories and LLM-chosen kinematic thresholds, but no evidence is given that this filtering has high recall; all downstream reasoning is conditioned on these timestamps.
  • domain assumption The perception outputs (PointPillars LiDAR detections, fine-tuned VLM object localization) are accurate enough for the reasoning agents.
    Section III-A says PointPillars plus clustering performs real-time LiDAR perception, and Module 2 assumes object labels and positions are reliable; degraded perception would propagate into the diagnostics.
  • domain assumption Ground-truth annotations for object detection and reasoning tasks are correct and complete.
    Section III-B defines the tasks and evaluation on the authors' own dataset; no inter-annotator agreement or external verification is reported, so label quality is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DriveAgent: Multi-Agent Structured Reasoning with LLM and Multimodal Sensor Fusion for Autonomous Driving." pith.science (2026). https://pith.science/paper/F3THU3Q5

@misc{pith2026250502123,
  author       = {Pith},
  title        = {Pith review of: DriveAgent: Multi-Agent Structured Reasoning with LLM and Multimodal Sensor Fusion for Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F3THU3Q5}},
  note         = {Machine review of arXiv:2505.02123}
}
read the original abstract

We introduce DriveAgent, a novel multi-agent autonomous driving framework that leverages large language model (LLM) reasoning combined with multimodal sensor fusion to enhance situational understanding and decision-making. DriveAgent uniquely integrates diverse sensor modalities-including camera, LiDAR, GPS, and IMU-with LLM-driven analytical processes structured across specialized agents. The framework operates through a modular agent-based pipeline comprising four principal modules: (i) a descriptive analysis agent identifying critical sensor data events based on filtered timestamps, (ii) dedicated vehicle-level analysis conducted by LiDAR and vision agents that collaboratively assess vehicle conditions and movements, (iii) environmental reasoning and causal analysis agents explaining contextual changes and their underlying mechanisms, and (iv) an urgency-aware decision-generation agent prioritizing insights and proposing timely maneuvers. This modular design empowers the LLM to effectively coordinate specialized perception and reasoning agents, delivering cohesive, interpretable insights into complex autonomous driving scenarios. Extensive experiments on challenging autonomous driving datasets demonstrate that DriveAgent is achieving superior performance on multiple metrics against baseline methods. These results validate the efficacy of the proposed LLM-driven multi-agent sensor fusion framework, underscoring its potential to substantially enhance the robustness and reliability of autonomous driving systems.

Figures

Figures reproduced from arXiv: 2505.02123 by the authors.

Figure 1
Figure 1. Overview of the inputs and outputs for the proposed DriveAgent [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An overview of the proposed architecture which is consisting of four modules (M1 to M4), where multimodal sensor inputs—camera, IMU, GPS [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Data collection vehicle sensor configuration and satellite images [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Overview of the multimodal reasoning pipeline used for driving scene understanding. Visual descriptions are generated from camera images, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Distribution of object categories in the human-annotated ground [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. A Survey on Vision-Language-Action Models for Autonomous Driving

    cs.CV 2025-06 conditional novelty 4.0 of 10

    A survey organizes vision-language-action models for autonomous driving into four stages, compares over 20 systems, and catalogs datasets, benchmarks, and open challenges.

  2. Chain-of-Thought for Autonomous Driving: A Comprehensive Survey and Future Prospects

    cs.RO 2025-05 conditional novelty 4.0 of 10

    A survey that classifies chain-of-thought methods for autonomous driving into modular, logical, and reflective pipelines, and proposes three evolutionary stages from direct prompting to reinforcement learning.

Reference graph

Works this paper leans on

26 extracted references · 10 canonical work pages · cited by 2 Pith papers

  1. [1]

    Autonomous driving’s future: convenient and connected,

    J. Deichmann, E. Ebel, K. Heineke, R. Heuss, M. Kellner, and F. Steiner, “Autonomous driving’s future: convenient and connected,” McKinsey & Company , 2023. [Online]. Avail- able: https://www.mckinsey.com/industries/automotive-and-assembly/ our-insights/autonomous-drivings-future-convenient-and-connected

    work page 2023

  2. [2]

    Large multimodal agents: A survey,

    J. Xie, Z. Chen, R. Zhang, X. Wan, and G. Li, “Large multimodal agents: A survey,” 2024. [Online]. Available: https: //arxiv.org/abs/2402.15116

    arXiv 2024

  3. [3]

    Llama: Open and efficient foundation language models,

    H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. Rozi`ere, N. Goyal, E. Hambro, F. Azhar et al., “Llama: Open and efficient foundation language models,” arXiv preprint arXiv:2302.13971, 2023

    arXiv 2023

  4. [4]

    Visual instruction tuning,

    H. Liu, C. Li, Q. Wu, and Y . J. Lee, “Visual instruction tuning,” Advances in neural information processing systems , vol. 36, pp. 34 892–34 916, 2023

    2023

  5. [5]

    A survey of reasoning with foundation models,

    J. Sun et al. , “A survey of reasoning with foundation models,” arXiv preprint arXiv:2401.12345 , 2024. [Online]. Available: https: //arxiv.org/abs/2401.12345

    arXiv 2024

  6. [6]

    Vision-language models for vision tasks: A survey,

    J. Zhang, J. Huang, S. Jin, and S. Lu, “Vision-language models for vision tasks: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

  7. [7]

    Drivelm: Driving with graph visual question answer- ing,

    C. Sima et al., “Drivelm: Driving with graph visual question answer- ing,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024

    work page 2024

  8. [8]

    V2v-llm: Vehicle-to-vehicle cooperative au- tonomous driving with multi-modal large language models,

    Y .-C. Chiu et al. , “V2v-llm: Vehicle-to-vehicle cooperative au- tonomous driving with multi-modal large language models,” arXiv preprint arXiv:2502.12345, 2025

    arXiv 2025

Show all 26 references
  1. [9]

    Genfollower: Enhancing car-following prediction with large language models,

    Z. Lan, H. Li, L. Liu, B. Fan, Y . Lv, Y . Ren, and Z. Cui, “Genfollower: Enhancing car-following prediction with large language models,” arXiv preprint arXiv:2407.05611, 2024. [Online]. Available: https://arxiv.org/abs/2407.05611

    2024 arXiv

  2. [10]

    Lmdrive: Closed-loop end-to-end driving with large language models,

    H. Shao, Y . Hu, L. Wang, S. L. Waslander, Y . Liu, and H. Li, “Lmdrive: Closed-loop end-to-end driving with large language models,” 2023. [Online]. Available: https://arxiv.org/abs/2312.07488

    2023 arXiv

  3. [11]

    Lampilot: An open benchmark dataset for autonomous driving with language model programs,

    Y . Ma, C. Cui, X. Cao, W. Ye, P. Liu, J. Lu, A. Abdelraouf, R. Gupta, K. Han, A. Bera, J. M. Rehg, and Z. Wang, “Lampilot: An open benchmark dataset for autonomous driving with language model programs,” in Proceedings of the Conference on Empirical Methods in Natural Language...

    2024

  4. [12]

    Koma: Knowledge-driven multi-agent framework for autonomous driving with large language models,

    K. Jiang, X. Cai, Z. Cui et al., “Koma: Knowledge-driven multi-agent framework for autonomous driving with large language models,” arXiv preprint arXiv:2407.14239, 2024

    2024 arXiv

  5. [13]

    Tree of thoughts: Deliberate problem solving with large language models,

    S. Yao, D. Yu, J. Zhao, I. Shafran, T. L. Griffiths, Y . Cao, and K. Narasimhan, “Tree of thoughts: Deliberate problem solving with large language models,” arXiv preprint arXiv:2305.10601 , 2023

    2023 arXiv

  6. [14]

    React: Synergizing reasoning and acting in language models,

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y . Cao, “React: Synergizing reasoning and acting in language models,” arXiv preprint arXiv:2210.03629, 2023

    2023 arXiv

  7. [15]

    Drivegpt4: Interpretable end-to-end autonomous driving via large language model. arxiv,

    Z. Xu, Y . Zhang, E. Xie, Z. Zhao, Y . Gou, Y . Kwan, K. Wong, Z. Li, and H. Zhao, “Drivegpt4: Interpretable end-to-end autonomous driving via large language model. arxiv,” arXiv preprint arXiv:2310.01412 , 2023

    2023 arXiv

  8. [16]

    Vlaad: Vision and language assistant for autonomous driving,

    S. Park, M. Lee, J. Kang, H. Choi, Y . Park, J. Cho, A. Lee, and D. Kim, “Vlaad: Vision and language assistant for autonomous driving,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 980–987

    2024

  9. [17]

    A mod- ular loop closure detection scheme for autonomous driving-a loosely coupled approach,

    W. Wang, H. Min, X. Wu, Y . Fang, G. Li, and X. Zhao, “A mod- ular loop closure detection scheme for autonomous driving-a loosely coupled approach,” IEEE Transactions on Vehicular Technology, 2024

    2024

  10. [18]

    Pointpillars: Fast encoders for object detection from point clouds,

    A. H. Lang, S. V ora, H. Caesar, L. Zhou, J. Yang, and O. Beijbom, “Pointpillars: Fast encoders for object detection from point clouds,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 12 697–12 705

    2019

  11. [19]

    Llama 3.2: Advancing vision ai on edge and mobile devices,

    M. AI, “Llama 3.2: Advancing vision ai on edge and mobile devices,” https://ai.meta.com/blog/ llama-3-2-connect-2024-vision-edge-mobile-devices/, 2024, accessed: 2025-04-12

    2024

  12. [20]

    Gpt-4o technical report,

    OpenAI, “Gpt-4o technical report,” 2024, https://openai.com/research/ gpt-4o

    2024

  13. [21]

    Pixtra: A large vision-language model,

    P. AI, “Pixtra: A large vision-language model,” 2024, https://pixtra.ai

    2024

  14. [22]

    Claude 3.7 sonnet model card,

    Anthropic, “Claude 3.7 sonnet model card,” 2024, https://www. anthropic.com/index/claude-3-7-sonnet

    2024

  15. [23]

    Lan- guage models are few-shot learners,

    T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhari- wal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell et al. , “Lan- guage models are few-shot learners,” Advances in neural information processing systems, vol. 33, pp. 1877–1901, 2020

    1901

  16. [24]

    Chain of thought prompting elicits reasoning in large language models,

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, B. Ichter, F. Xia, E. Chi, Q. V . Le, and D. Zhou, “Chain of thought prompting elicits reasoning in large language models,” arXiv preprint arXiv:2201.11903 , 2022

    2022 arXiv

  17. [25]

    Self-refine: Iterative refinement with self-feedback,

    A. Madaan, S. Lin, X. Liu, D. Zhou, Q. V . Le, D. Schuurmans, E. H. Chi, and J. Wei, “Self-refine: Iterative refinement with self-feedback,” arXiv preprint arXiv:2303.17651 , 2023

    2023 arXiv

  18. [26]

    Lora: Low-rank adaptation of large language models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chen et al., “Lora: Low-rank adaptation of large language models.” ICLR, vol. 1, no. 2, p. 3, 2022

    2022

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.