pith. sign in

arxiv: 2512.17321 · v2 · submitted 2025-12-19 · 💻 cs.RO

Neuro-Symbolic Control with Large Language Models for Language-Guided Spatial Tasks

Momina Liaqat Ali , Muhammad Abid , Muhammad Saqlain , Jose M. Merigo This is my paper

Pith reviewed 2026-05-16 21:16 UTC · model grok-4.3

classification 💻 cs.RO
keywords neuro-symbolic controllarge language modelsspatial manipulationlanguage-guided roboticsdelta controllerembodied AIcontinuous controlsymbolic reasoning
0
0 comments X

The pith

A neuro-symbolic framework lets an LLM handle high-level language reasoning while a neural delta controller executes precise continuous motions, raising success rates and cutting steps by more than 70 percent in spatial manipulation tasks.

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

The paper tests whether splitting control between a language model for symbolic task interpretation and a lightweight neural controller for incremental physical actions improves performance over using either component alone. In planar object-manipulation experiments defined by spatial language instructions, the combined system is compared against pure LLM baselines and pure neural baselines across several locally run models. Results show higher task completion rates together with average step reductions above 70 percent and speedups reaching 8.83 times, all without reinforcement learning or environment rollouts. The neural controller is trained once on synthetic geometric data and then used unchanged, which the authors credit for stability and robustness to changes in language-model quality. This decomposition is presented as a practical route to reliable language-guided embodied control.

Core claim

The central claim is that assigning symbolic task interpretation to a locally deployed LLM while routing uninterpreted low-level execution to a neural delta controller trained on artificial geometric data produces higher success rates, more than 70 percent fewer steps on average, and speedups up to 8.83 times compared with LLM-only or neural-only baselines in language-specified planar manipulation tasks.

What carries the argument

The neuro-symbolic split that keeps the LLM on symbolic outputs and delegates bounded incremental actions in continuous space to a neural delta controller trained only on synthetic geometry.

If this is right

  • Success rates rise because the LLM is prevented from generating hallucinated low-level actions.
  • Step counts and execution time drop sharply once motion execution is off-loaded to the trained delta controller.
  • Performance stays consistent across different language models because the neural component absorbs the continuous-control burden.
  • No reinforcement learning or costly rollouts are required, lowering the barrier to deployment.
  • Interpretability increases because the LLM's output remains symbolic and inspectable.
  • pith_inferences=[

Load-bearing premise

A lightweight neural delta controller trained solely on artificial geometric data can execute the required bounded incremental actions reliably in the target continuous space without further training or adaptation.

What would settle it

Running the same tasks in a new continuous-space geometry where the neural controller produces visibly incorrect incremental moves or the overall success rate falls below the LLM-only baseline would falsify the claim that the split reliably improves performance.

Figures

Figures reproduced from arXiv: 2512.17321 by Jose M. Merigo, Momina Liaqat Ali, Muhammad Abid, Muhammad Saqlain.

Figure 1
Figure 1. Figure 1: Motivation for neuro-symbolic control. End-to-end LLM-based control directly predicts continuous actions, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed neuro-symbolic control framework. A local large language model performs [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Success rate aggregated across all language models for each spatial task. The proposed LLM+DL framework [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Total average number of control steps for all language models. Compared to LLM-only control, the LLM+DL [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Normalized distance-to-goal over time for the [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Speedup of LLM+DL relative to LLM-only control aggregated across language models. The proposed [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Success rate by language model and task. The suggested LLM+DL framework consistently improves [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Success rate by task and language model. The neuro-symbolic LLM+DL framework improves reliability [PITH_FULL_IMAGE:figures/full_fig_p020_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The average number of control steps for the [PITH_FULL_IMAGE:figures/full_fig_p021_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Normalized distance-to-goal over time for all spatial tasks. The proposed LLM+DL framework converges [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Model-specific speedup (left) and success-rate improvement (right) of LLM+DL over LLM-only control. [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
read the original abstract

Although large language models (LLMs) have recently become effective tools for language-conditioned control in embodied systems, instability, slow convergence, and hallucinated actions continue to limit their direct application to continuous control. A modular neuro-symbolic control framework that clearly distinguishes between low-level motion execution and high-level semantic reasoning is proposed in this work. While a lightweight neural delta controller performs bounded, incremental actions in continuous space, a locally deployed LLM interprets symbolic tasks. We assess the suggested method in a planar manipulation setting with spatial relations between objects specified by language. Numerous tasks and local language models, such as Mistral, Phi, and LLaMA-3.2, are used in extensive experiments to compare LLM-only control, neural-only control, and the suggested LLM+DL framework. In comparison to LLM-only baselines, the results show that the neuro-symbolic integration consistently increases both success rate and efficiency, achieving average step reductions exceeding 70% and speedups of up to 8.83x while remaining robust to language model quality. The suggested framework enhances interpretability, stability, and generalization without any need of reinforcement learning or costly rollouts by controlling the LLM to symbolic outputs and allocating uninterpreted execution to a neural controller trained on artificial geometric data. These outputs show empirically that neuro-symbolic decomposition offers a scalable and principled way to integrate language understanding with ongoing control, this approach promotes the creation of dependable and effective language-guided embodied systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a modular neuro-symbolic framework for language-guided planar manipulation tasks. A locally deployed LLM handles high-level symbolic reasoning and task interpretation, while a lightweight neural delta controller (trained exclusively on synthetic geometric data) executes bounded incremental actions in continuous space. Experiments compare LLM-only, neural-only, and combined LLM+DL approaches across multiple spatial-relation tasks and models (Mistral, Phi, LLaMA-3.2), reporting consistent gains in success rate and efficiency: average step reductions exceeding 70% and speedups up to 8.83x, with robustness to LLM quality and no requirement for RL or rollouts.

Significance. If the empirical claims hold under rigorous validation, the work demonstrates a practical, interpretable decomposition that mitigates LLM instability in continuous control while avoiding expensive training. This could offer a scalable template for integrating language understanding with low-level execution in embodied systems, particularly where direct LLM control or end-to-end RL is impractical.

major comments (3)
  1. [Abstract / Experiments] Abstract and Experiments section: The central quantitative claims (average step reductions >70%, speedups up to 8.83x, robustness across Mistral/Phi/LLaMA-3.2) are presented without task definitions, number of trials per condition, statistical tests, error bars, or variance measures. These omissions make it impossible to evaluate whether the reported efficiency gains are statistically reliable or merely anecdotal.
  2. [Controller / Experiments] Controller description (likely §3–4): The neural delta controller is trained only on artificial geometric data and asserted to execute bounded incremental actions reliably in continuous planar space. No ablation isolates controller error under task-relevant conditions (object contact, friction, sensor noise, non-convex geometries), nor is any quantitative bound on cumulative position drift after k steps provided. This assumption is load-bearing for the claimed 70%+ step reductions; if drift exceeds symbolic-layer tolerance, the neuro-symbolic gains collapse regardless of LLM quality.
  3. [Experiments / Baselines] Baseline comparison: The LLM-only baseline is described as directly outputting actions, yet the precise prompting strategy, action discretization, and failure modes (hallucinations, instability) are not detailed. Without this, it is unclear whether the reported improvements stem from the neuro-symbolic split or from differences in how the LLM is constrained to symbolic outputs.
minor comments (2)
  1. [Abstract] Abstract phrasing: 'Numerous tasks and local language models, such as Mistral, Phi, and LLaMA-3.2, are used in extensive experiments' is grammatically awkward and should be rephrased for clarity (e.g., 'extensive experiments are conducted across numerous tasks using local LLMs including...').
  2. [Abstract / Conclusion] The claim of 'no need of reinforcement learning or costly rollouts' is repeated; a single concise statement in the introduction or conclusion would suffice.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, indicating the revisions that will be incorporated into the next manuscript version.

read point-by-point responses

  1. Referee: [Abstract / Experiments] Abstract and Experiments section: The central quantitative claims (average step reductions >70%, speedups up to 8.83x, robustness across Mistral/Phi/LLaMA-3.2) are presented without task definitions, number of trials per condition, statistical tests, error bars, or variance measures. These omissions make it impossible to evaluate whether the reported efficiency gains are statistically reliable or merely anecdotal.

    Authors: We agree that these experimental details are necessary for rigorous evaluation. In the revised manuscript we will add explicit definitions of all spatial-relation tasks (including a summary table), the exact number of trials per condition (50 independent trials for each task-model pair), standard-deviation error bars on all reported metrics, and statistical significance tests (paired Wilcoxon signed-rank tests) comparing the neuro-symbolic method against baselines. These additions will appear in both the abstract and the Experiments section. revision: yes

  2. Referee: [Controller / Experiments] Controller description (likely §3–4): The neural delta controller is trained only on artificial geometric data and asserted to execute bounded incremental actions reliably in continuous planar space. No ablation isolates controller error under task-relevant conditions (object contact, friction, sensor noise, non-convex geometries), nor is any quantitative bound on cumulative position drift after k steps provided. This assumption is load-bearing for the claimed 70%+ step reductions; if drift exceeds symbolic-layer tolerance, the neuro-symbolic gains collapse regardless of LLM quality.

    Authors: The referee correctly highlights a missing robustness analysis. While the bounded-action design was intended to limit drift, we did not quantify controller error under realistic conditions. In the revision we will add an ablation evaluating the delta controller under simulated sensor noise, friction, and contact scenarios, together with an analytic upper bound on cumulative position drift after k steps derived from the action bounds and measured controller accuracy on the synthetic data. These results will be presented in a new subsection of the Experiments section. revision: yes

  3. Referee: [Experiments / Baselines] Baseline comparison: The LLM-only baseline is described as directly outputting actions, yet the precise prompting strategy, action discretization, and failure modes (hallucinations, instability) are not detailed. Without this, it is unclear whether the reported improvements stem from the neuro-symbolic split or from differences in how the LLM is constrained to symbolic outputs.

    Authors: We agree that the LLM-only baseline description is insufficiently detailed. The revised manuscript will expand this section to include the exact prompt template used to elicit direct continuous actions, the discretization scheme (fixed increments in x, y, and rotation), and the observed failure modes (action hallucinations, instability over long horizons). This clarification will demonstrate that the performance gains arise from the neuro-symbolic decomposition rather than prompting differences alone. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical comparisons rest on independent evaluations

full rationale

The paper reports direct experimental results comparing LLM-only, neural-only, and neuro-symbolic control across multiple spatial tasks and local LLMs (Mistral, Phi, LLaMA-3.2). Central metrics (success rate, >70% step reduction, up to 8.83x speedup) are obtained from explicit trials rather than any derivation, equation, or fitted parameter that reduces to its own inputs by construction. The neural delta controller is trained on separate artificial geometric data and evaluated as an independent module; no self-citation chain, ansatz smuggling, or uniqueness theorem is invoked to justify the performance claims. The framework is self-contained against external benchmarks via ablation-style comparisons, yielding no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the unproven effectiveness of the neural delta controller for continuous execution and the assumption that symbolic outputs from the LLM map cleanly to actionable increments without further verification.

axioms (1)
  • domain assumption Lightweight neural delta controller trained on artificial geometric data suffices for bounded continuous control in the target domain.
    Invoked to justify allocation of execution to the neural component without RL.

pith-pipeline@v0.9.0 · 5572 in / 1182 out tokens · 37899 ms · 2026-05-16T21:16:45.265410+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages · 4 internal anchors

  1. [1]

    Bandyopadhyay, D., Bhattacharjee, S., & Ekbal, A. (2025). Thinking machines: A survey of LLM-based reasoning strategies.arXiv preprint arXiv:2503.10814

    work page arXiv 2025

  2. [2]

    Zhang, Y ., Wang, H., Feng, S., Tan, Z., Han, X., He, T., & Tsvetkov, Y . (2024). Can LLM graph reasoning generalize beyond pattern memorization?arXiv preprint arXiv:2406.15992

    work page arXiv 2024

  3. [3]

    Liu, L., Nair, A., Peng, T., Desai, S., Gupta, M., Mehta, K., & Singh, P. (2024). Optimizing task planning efficiency in LLMs: Beyond closed-loop systems.Authorea Preprints

    work page 2024

  4. [4]

    Banerjee, S., Agarwal, A., & Singla, S. (2025). LLMs will always hallucinate, and we need to live with this. In Proceedings of the Intelligent Systems Conference(pp. 624–648). Springer

    work page 2025

  5. [5]

    Huang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H., & Liu, T. (2025). A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions.ACM Transactions on Information Systems, 43(2), 1–55

    work page 2025

  6. [6]

    Tang, C., Abbatematteo, B., Hu, J., Chandra, R., Martín-Martín, R., & Stone, P. (2025). Deep reinforcement learning for robotics: A survey of real-world successes.Annual Review of Control, Robotics, and Autonomous Systems, 8, 153–178

    work page 2025

  7. [7]

    Du, Q., Li, B., Du, Y ., Su, S., Fu, T., Zhan, Z., & Wang, C. (2025). Fast task planning with neuro-symbolic relaxation.arXiv preprint arXiv:2507.15975

    work page arXiv 2025

  8. [8]

    Grounding llms for robot task planning using closed-loop state feedback,

    Bhat, V ., Kaypak, A. U., Krishnamurthy, P., Karri, R., & Khorrami, F. (2024). Grounding LLMs for robot task planning using closed-loop state feedback.arXiv preprint arXiv:2402.08546

    work page arXiv 2024

  9. [9]

    Su, W. (2025). Do large language models (really) need statistical foundations?arXiv preprint arXiv:2505.19145

    work page arXiv 2025

  10. [10]

    Enoasmo, V ., Featherstonehaugh, C., Konstantinopoulos, X., & Huntington, Z. (2025). Structural embedding projection for contextual large language model inference.arXiv preprint arXiv:2501.18826

    work page arXiv 2025

  11. [11]

    (2022, October)

    Ullah, S., Liaqat, M., Asif, A., Khan, A., Aslam, U., & Asif, H. (2022, October). Deep auto encoder based chatbot for discrete math course. In 2022 International Conference onRecent Advances in Electrical Engineering & Computer Sciences (RAEE & CS)(pp. 1-7). IEEE

    work page 2022

  12. [12]

    Kim, Y ., Choi, J., & Lee, S. (2024). A survey on integration of large language models with intelligent robots. Intelligent Service Robotics

    work page 2024

  13. [13]

    Zeng, F., Gan, W., Wang, Y ., Liu, N., & Yu, P. S. (2023). Large language models for robotics: A survey.arXiv preprint arXiv:2311.07226

    work page arXiv 2023

  14. [14]

    Huang, W., Abbeel, P., Pathak, D., & Mordatch, I. (2022). Language models as zero-shot planners: Extracting actionable knowledge for embodied agents. InProceedings of ICML

    work page 2022

  15. [15]

    Ahn, M., Brohan, A., Brown, N., Chebotar, Y ., Cortes, O., David, B., & Zeng, A. (2022). Do as I can, not as I say: Grounding language in robotic affordances.arXiv preprint arXiv:2204.01691

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  16. [16]

    S., Lynch, C., Chowdhery, A., Wahid, A., & Florence, P

    Driess, D., Xia, F., Sajjadi, M. S., Lynch, C., Chowdhery, A., Wahid, A., & Florence, P. (2023). PaLM-E: An embodied multimodal language model

    work page 2023

  17. [17]

    Liang, J., Huang, W., Xia, F., Xu, P., Hausman, K., Ichter, B., & Zeng, A. (2022). Code as policies: Language model programs for embodied control.arXiv preprint arXiv:2209.07753

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  18. [19]

    Jeong, H., Lee, H., Kim, C., & Shin, S. (2024). A survey of robot intelligence with large language models.Applied Sciences, 14(19), 8868. 17 Running Title for Header

    work page 2024

  19. [20]

    Abid, M., Akhtar, T., & Bhatt, H. (2025). Uncertainty quantification in steady-state heat transfer: A comprehensive analysis of DRAM and MCMC methods with applications to thermal systems.Spectrum of engineering and management sciences, 3(1), 63-75

    work page 2025

  20. [21]

    Chen, Y ., Arkin, J., Zhang, Y ., et al. (2024). AutoTAMP: Autoregressive task and motion planning with LLMs as translators and checkers. InProceedings of ICRA

    work page 2024

  21. [22]

    Garcez, A., & Lamb, L. (2023). Neurosymbolic AI: The third wave.Artificial Intelligence Review, 56, 12387–12406

    work page 2023

  22. [23]

    arXiv preprint arXiv:2401.01040 , year=

    Wan, Z., Liu, C. K., Yang, H., Li, C., You, H., Fu, Y ., & Raychowdhury, A. (2024). Towards cognitive AI systems: A survey and prospective on neuro-symbolic AI.arXiv preprint arXiv:2401.01040

    work page arXiv 2024

  23. [24]

    De Raedt, L., Dumancic, S., Manhaeve, R., & Marra, G. (2024). From statistical relational to neurosymbolic artificial intelligence: A survey.Artificial Intelligence, 328

    work page 2024

  24. [25]

    Neuro-symbolic ai in 2024: A systematic review.arXiv preprint arXiv:2501.05435, 2025

    Colelough, B. C., & Regli, W. (2025). Neuro-symbolic AI in 2024: A systematic review.arXiv preprint arXiv:2501.05435

    work page arXiv 2025

  25. [26]

    Abid, M., & Saqlain, M. (2024). Optimizing diabetes data insights through kmapper-based topological networks: a decision analytics approach for predictive and prescriptive modeling.Management science advances, 1(1), 1-19

    work page 2024

  26. [27]

    R., Chitnis, R., Holladay, R

    Garrett, C. R., Chitnis, R., Holladay, R. M., et al. (2021). Integrated task and motion planning.Annual Review of Control, Robotics, and Autonomous Systems, 4, 265–293

    work page 2021

  27. [28]

    Zhai, W., Liao, J., Chen, Z., Su, B., & Zhao, X. (2025). A survey of task planning with large language models. Intelligent Computing, 4, 0124

    work page 2025

  28. [29]

    Bousetouane, B. (2025). Agentic LLM-based robotic systems for real-world applications: A review.Frontiers in Robotics and AI

    work page 2025

  29. [30]

    P., Brundage, M., & Bharath, A

    Arulkumaran, K., Deisenroth, M. P., Brundage, M., & Bharath, A. A. (2017). Deep reinforcement learning: A brief survey.IEEE Signal Processing Magazine, 34(6), 26–38

    work page 2017

  30. [31]

    M., Thuong, L

    Thanh, L. M., Thuong, L. H., Loc, P. T., & Nguyen, C.-N. (2020). Delta robot control using single neuron PID algorithms based on recurrent fuzzy neural network identifiers.International Journal of Mechanical Engineering and Robotics Research, 9(10), 1411–1418

    work page 2020

  31. [32]

    Gholami, A., Homayouni, T., Ehsani, R., & Sun, J. Q. (2021). Inverse kinematic control of a delta robot using neural networks in real-time.Robotics, 10(4), 115

    work page 2021

  32. [33]

    Fan, Y ., Huang, H., & Yang, C. (2022). Fixed-time incremental neural control for manipulator based on composite learning with input saturation. InActuators, 11(12), 373

    work page 2022

  33. [34]

    A., Steinmetz, R., & Tello Gamarra, D

    dos Santos Lima, M., Kich, V . A., Steinmetz, R., & Tello Gamarra, D. F. (2024). Delta robot control by learning systems: Harnessing the power of deep reinforcement learning algorithms.Journal of Intelligent & Fuzzy Systems, 46(2), 4881–4894

    work page 2024

  34. [35]

    Abid, M., & Ali, M. L. Enhancing Software Effort Estimation: A Comparative Analysis of Machine Learning Models with Correlation-Based Feature Selection.Sustainable Machine Intelligence Journal, 12, 1-17

    work page

  35. [36]

    Khosravi, S., & Akbari, A. (2022). Experimental study on a novel simultaneous control and identification of a 3-DOF delta robot using model reference adaptive control.Mechatronics, 86

    work page 2022

  36. [37]

    Chen, B., Xu, Z., Kirmani, S., et al. (2024). SpatialVLM: Endowing vision-language models with spatial reasoning capabilities. InProceedings of CVPR

    work page 2024

  37. [38]

    Rana, K., Haviland, J., Garg, S., et al. (2023). SayPlan: Grounding large language models using 3D scene graphs for scalable robot task planning. InProceedings of CoRL

    work page 2023

  38. [39]

    Ramchurn, and Mohammad D

    Hunt, W., Ramchurn, S. D., & Soorati, M. D. (2024). A survey of language-based communication in robotics. arXiv preprint arXiv:2406.04086

    work page arXiv 2024

  39. [40]

    Abid, M., Bukhari, S., & Saqlain, M. (2025). Enhancing software effort Estimation in healthcare informatics: A comparative analysis of machine learning models with Correlation-Based feature selection.Sustainable Machine Intelligence Journal, 10, 50-66

    work page 2025

  40. [41]

    Wang, J., Shi, E., Hu, H., Ma, C., Liu, Y ., Wang, X., & Zhang, S. (2024). Large language models for robotics: Opportunities, challenges, and perspectives.Journal of Automation and Intelligence

    work page 2024

  41. [42]

    (2024, October)

    Amin, B. (2024, October). Mistral expands its reach in the SLM space with Ministral models.TechTalks

    work page 2024

  42. [43]

    Zheng, Y ., Chen, Y ., Qian, B., Shi, X., Shu, Y ., & Chen, J. (2025). A review on edge large language models: Design, execution, and applications.ACM Computing Surveys, 57(8), 1–35. 18 Running Title for Header

    work page 2025

  43. [44]

    From clip to dino: Visual encoders shout in multi-modal large language models,

    Jiang, D., Liu, Y ., Liu, S., Zhao, J. E., Zhang, H., Gao, Z., & Xiong, H. (2023). From CLIP to DINO: Visual encoders shout in multi-modal large language models.arXiv preprint arXiv:2310.08825

    work page arXiv 2023

  44. [45]

    Abdin, M., Aneja, J., Behl, H., Bubeck, S., Eldan, R., Gunasekar, S., & Zhang, Y . (2024). Phi-4 technical report. arXiv preprint arXiv:2412.08905

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  45. [46]

    LLaMA: Open and Efficient Foundation Language Models

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M. A., Lacroix, T., & Lample, G. (2023). LLaMA: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  46. [47]

    Data Science Dojo. (2024). Phi-3 and beyond: Top small language models of 2024

    work page 2024

  47. [48]

    (2024, October)

    Mistral AI. (2024, October). Introducing Les Ministraux: Edge-optimized models

    work page 2024

  48. [49]

    Kress-Gazit, H., Hashimoto, K., Kuppuswamy, N., Shah, P., Horgan, P., Richardson, G., & Burchfiel, B. (2024). Robot learning as an empirical science: Best practices for policy evaluation.arXiv preprint arXiv:2409.09491

    work page arXiv 2024

  49. [50]

    Faigl, J., Kulich, M., & Pˇreuˇcil, L. (2012). Goal assignment using distance cost in multi-robot exploration. In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)(pp. 3741–3746). 19 Running Title for Header Supplementary Material Additional experimental data and analysis that corroborate the conclusions in the ...

    work page 2012