pith. machine review for the scientific record. sign in

arxiv: 2605.09146 · v1 · submitted 2026-05-09 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

Beyond Thinking: Imagining in 360^circ for Humanoid Visual Search

Jingdong Zhang, Wenping Wang, Xiaohang Zhan, Xin Li, Yizhou Wang, Zhengzhong Tu

Pith reviewed 2026-05-12 02:48 UTC · model grok-4.3

classification 💻 cs.CV
keywords humanoid visual search360 degree explorationsemantic spatial priorsimaginator actor frameworkactive visual navigationembodied AIprobabilistic layout predictiondecoupled reasoning
0
0 comments X

The pith

A single-step probabilistic predictor of semantic layouts lets humanoids search 360° scenes efficiently without building cumulative reasoning chains.

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

The paper introduces a decoupled framework for humanoid visual search that splits the task between an Imaginator and an Actor. The Imaginator predicts the semantic layout of both seen and unseen regions in one probabilistic step and supplies the Actor with multiple sampled hypotheses rather than a growing chain of thoughts. This design removes the need for expensive full-trajectory Chain-of-Thought annotations and produces over 1.96 million training examples. Experiments show the approach raises search success rates and cuts steps taken in complex real-world environments.

Core claim

By replacing cumulative multi-turn reasoning with a single-step probabilistic Imaginator that infers semantic spatial priors for observed and unobserved areas and supplies a distribution of hypotheses to the Actor, the framework supplies robust guidance that hedges uncertainty, eliminates trajectory-level annotations, and yields large training sets that improve efficiency and success rates in in-the-wild 360° search.

What carries the argument

The Imaginator, a probabilistic model that predicts full semantic spatial layouts of both observed and unobserved regions in one step and samples multiple hypotheses to guide the Actor.

If this is right

  • Full-trajectory Chain-of-Thought annotations are no longer required, allowing generation of over 1.96 million curated training samples.
  • Multiple sampled hypotheses from the semantic prior hedge uncertainty and reduce dead-end explorations.
  • Search efficiency and success rates rise in complex, in-the-wild 360° environments compared with monolithic reasoning approaches.
  • The cognitive load of maintaining long reasoning chains is removed from the action policy.

Where Pith is reading between the lines

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

  • The same single-step prior prediction could be tested in other embodied tasks such as object manipulation or long-range navigation where full-scene semantics matter.
  • Training data volume could grow further if the Imaginator is trained on synthetic 360° renderings rather than real trajectories.
  • In dynamic scenes the framework might need an update rule that refreshes the prior after each new observation while keeping the single-step structure.

Load-bearing premise

That a single probabilistic prediction of semantic layouts can supply enough robust guidance to hedge uncertainty without needing iterative multi-turn reasoning.

What would settle it

A head-to-head test in environments with high visual ambiguity or sudden changes where the single-step predictions produce lower success rates or longer paths than cumulative-reasoning baselines.

Figures

Figures reproduced from arXiv: 2605.09146 by Jingdong Zhang, Wenping Wang, Xiaohang Zhan, Xin Li, Yizhou Wang, Zhengzhong Tu.

Figure 1
Figure 1. Figure 1: Imagining in 360◦ : We propose a decoupled architecture for Humanoid Visual Search (HVS). The Imaginator explicitly models the 360◦ environment by predicting the semantic layout of both observed and unseen regions, which provides the downstream Actor with a sampled distribution of spatial priors as suggestions. This explicit imagination empowers the agent to reorient its head more efficiently, hedging agai… view at source ↗
Figure 2
Figure 2. Figure 2: Framework Overview. (a) 360◦ Env Imagination: The Imaginator performs a single-step estimation of the global layout by predicting (φ, µ) coordinates for both observed landmarks and imagined entities to capture high-level spatial relationships. (b) Collaborative HVS Pipeline: The Actor integrates the sampled distribution of probabilistic spatial suggestions into its reasoning chain, allowing the agent to he… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the Curated Training Dataset. (a) Distribution of scene categories, highlight￾ing a diverse mix dominated by most commonly encountered outdoor urban (31.6%) and residential interior (25.9%) environments. (b) The composition of unlabeled panorama sources. (c) The dis￾tribution of labeled semantic item categories. (d) The histogram of labeled items per panorama, with a median of 11 items, demonst… view at source ↗
Figure 4
Figure 4. Figure 4: Scaling Laws of Imagination. We use Qwen2.5-VL-3B as the Actor model [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Distribution Analysis. (a) Histograms of the step distribution, illustrating the percentage of search episodes that terminate at each step for both the baseline (Actor Only) and our joint pipeline across the HOS and HPS tasks. (b) Step-by-step heatmap visualizations of the Imaginator’s probabilistic spatial suggestions. The heat areas indicate the probability distribution of the imagined target coordinates… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative Search Trajectories in the wild. Evaluated in complex environments generated by Gemini 3.1-Pro. The Actor loses context and falls into an infinite dead loop, whereas our decoupled framework leverages layout imagination to locate the target within 2-3 efficient steps. single-step sampling lacks sequential reasoning. Yet, integrating it with the Actor unlocks massive gains, proving our framework … view at source ↗
Figure 7
Figure 7. Figure 7: Overview of the Curated Training Dataset. (a) Fine-grained scene category distribution. (b) Distribution of target-search instructions. (c-e) Semantic category distributions for Imagined entities, Observed entities, and Search Targets, respectively. • Stage 2 (Clean-data SFT): The pseudo-pretrained checkpoints are fine-tuned on the curated H*Bench set (∼4,248 samples) for 2 epochs to compensate for the sma… view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative search trajectories in synthetic panoramas. Evaluated in complex indoor and outdoor environments generated by GPT-Image and Gemini 3.1-Pro. The step-by-step imagination distributions (heatmaps) demonstrate how the framework progressively refines its spatial hypotheses as new visual evidence is gathered, effectively guiding both HOS and HPS tasks. A.3 More Quantitative Comparisons To provide a c… view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative search trajectories in real-world panoramas. Evaluated in diverse, in￾the-wild Google Street View environments. The decoupled pipeline demonstrates strong zero-shot robustness in unstructured outdoor scenes (e.g., ski resorts, bustling city streets, and suburban neighborhoods), efficiently narrowing down target locations. Efficacy on State-of-the-Art Proprietary Models. While cutting-edge propr… view at source ↗
read the original abstract

Humanoid Visual Search (HVS) requires agents to actively explore immersive 360$^\circ$ environments. While prior methods treat this as a monolithic task relying on cumulative, multi-turn Chain-of-Thought (CoT) reasoning, they impose heavy cognitive burdens and require expensive trajectory-level annotations. In this paper, we propose Imagining in 360$^\circ$, a novel framework that decouples the exploration process into a specialized Imaginator and an Actor. The Imaginator functions as a probabilistic predictor of spatial priors; instead of maintaining a cumulative reasoning chain, it infers the semantic layout of both observed and unobserved regions in a single step. By sampling multiple hypotheses within this semantic space, we provide the Actor with a distribution of effective spatial information, offering robust guidance that hedges against uncertainty during active search. This decoupled architecture significantly lowers data engineering costs by eliminating the need for full-trajectory CoT annotations, enabling the generation of over 1.96 million curated training samples. Extensive experiments demonstrate that explicitly modeling semantic spatial priors drastically improves search efficiency and success rates in complex, in-the-wild environments.

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 paper proposes 'Imagining in 360°', a decoupled framework for Humanoid Visual Search (HVS) in immersive 360° environments. It introduces an Imaginator module that performs single-step probabilistic prediction of semantic spatial layouts over both observed and unobserved regions, sampling multiple hypotheses to guide an Actor module. This replaces cumulative multi-turn Chain-of-Thought reasoning, reduces annotation costs, and enables creation of a 1.96 million sample dataset. The central claim is that explicitly modeling these semantic priors yields substantial gains in search efficiency and success rates in complex in-the-wild settings.

Significance. If the experimental claims hold with proper controls, the work could meaningfully advance efficient active perception for humanoid agents by shifting from heavy reasoning chains to prior-based hypothesis sampling. The scale of the curated dataset is a concrete strength that may support future research in robotics and embodied AI.

major comments (3)
  1. [Abstract] Abstract: the assertion of 'drastically improves search efficiency and success rates' is load-bearing for the contribution yet supplies no baselines, metrics (e.g., success rate, steps-to-goal), error bars, or statistical controls, leaving the magnitude and reliability of the improvement impossible to evaluate from the provided text.
  2. [Method] Method description (Imaginator): the claim that single-step probabilistic output over observed+unobserved semantics supplies the Actor with a distribution 'sufficient for efficient search' and 'hedges against uncertainty' without multi-turn accumulation is central, but the manuscript provides no analysis of prediction error rates on unobserved regions, how multiple hypothesis sampling bounds compounding errors, or failure cases when the prior is inaccurate.
  3. [Experiments] Experiments: no comparison is shown to strong multi-turn CoT baselines or ablations that isolate the contribution of the Imaginator versus the Actor, which is required to substantiate the decoupling benefit and the dataset-size advantage.
minor comments (2)
  1. [Abstract] Abstract: 'in-the-wild environments' should be accompanied by concrete dataset statistics or scene diversity metrics to clarify the evaluation scope.
  2. [Method] Notation: the distinction between 'semantic layout' and 'spatial priors' is used interchangeably in places; a short clarifying sentence or diagram would improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

Thank you for the constructive feedback on our manuscript. We address each major comment point by point below. We have revised the manuscript to strengthen the presentation of results and analysis where the comments identify gaps.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion of 'drastically improves search efficiency and success rates' is load-bearing for the contribution yet supplies no baselines, metrics (e.g., success rate, steps-to-goal), error bars, or statistical controls, leaving the magnitude and reliability of the improvement impossible to evaluate from the provided text.

    Authors: We agree that the abstract should convey concrete quantitative support for the central claim. The full experimental section already reports success rates, steps-to-goal, baselines, error bars, and statistical controls. We have revised the abstract to explicitly reference these metrics (e.g., relative success-rate gains and efficiency reductions) while remaining within length limits, directing readers to the detailed tables and controls in the experiments. revision: yes

  2. Referee: [Method] Method description (Imaginator): the claim that single-step probabilistic output over observed+unobserved semantics supplies the Actor with a distribution 'sufficient for efficient search' and 'hedges against uncertainty' without multi-turn accumulation is central, but the manuscript provides no analysis of prediction error rates on unobserved regions, how multiple hypothesis sampling bounds compounding errors, or failure cases when the prior is inaccurate.

    Authors: This observation correctly identifies an opportunity to make the Imaginator's robustness more explicit. Although end-to-end task performance already demonstrates the practical value of the single-step prior, we have added a dedicated analysis subsection quantifying Imaginator error rates on unobserved regions, showing how multi-hypothesis sampling limits error propagation, and including representative failure cases with discussion of when the semantic prior is inaccurate. revision: yes

  3. Referee: [Experiments] Experiments: no comparison is shown to strong multi-turn CoT baselines or ablations that isolate the contribution of the Imaginator versus the Actor, which is required to substantiate the decoupling benefit and the dataset-size advantage.

    Authors: We accept that direct isolation of the decoupling benefit strengthens the claims. The original experiments compared against prior HVS methods, but we have now added (i) strong multi-turn CoT baselines adapted to the 360° setting and (ii) ablations that disable or replace the Imaginator while keeping the Actor fixed. These additions quantify the contribution of the single-step prior and the annotation-efficiency advantage that enabled the 1.96 M sample dataset. revision: yes

Circularity Check

0 steps flagged

No circularity in architectural proposal or claims

full rationale

The paper introduces a decoupled Imaginator-Actor framework for humanoid visual search, where the Imaginator performs single-step probabilistic semantic layout prediction over observed and unobserved regions, and multiple hypotheses are sampled to guide the Actor. No equations, derivations, or parameter-fitting steps are described that reduce outputs to inputs by construction. The architecture is presented as a novel design choice that eliminates the need for trajectory-level CoT annotations, enabling large-scale data generation as a downstream benefit rather than a fitted input. No self-citations are invoked as load-bearing uniqueness theorems, and no ansatz or renaming of known results occurs. The central claim of improved efficiency rests on experimental validation in in-the-wild environments, not on self-referential definitions. This is a self-contained architectural contribution with no detectable circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Abstract-only review provides no equations or implementation details, so free parameters, axioms, and invented entities cannot be enumerated beyond the high-level modules named in the text.

invented entities (2)
  • Imaginator no independent evidence
    purpose: Probabilistic single-step predictor of semantic layouts for observed and unobserved regions
    New module introduced to replace cumulative CoT reasoning; no independent evidence outside the paper is provided.
  • Actor no independent evidence
    purpose: Uses sampled hypotheses from Imaginator to guide active search
    Paired module whose behavior depends on the new Imaginator output.

pith-pipeline@v0.9.0 · 5510 in / 1265 out tokens · 43858 ms · 2026-05-12T02:48:41.077518+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

115 extracted references · 115 canonical work pages · 17 internal anchors

  1. [1]

    2010 , publisher=

    Vision: A computational investigation into the human representation and processing of visual information , author=. 2010 , publisher=

    work page 2010

  2. [2]

    2025 , booktitle=

    SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training , author=. 2025 , booktitle=

    work page 2025

  3. [3]

    arXiv preprint arXiv:2505.08243 , year=

    Training Strategies for Efficient Embodied Reasoning , author=. arXiv preprint arXiv:2505.08243 , year=

    work page arXiv

  4. [4]

    Science , volume=

    Eye and head movements in peripheral vision: nature of compensatory eye movements , author=. Science , volume=. 1966 , publisher=

    work page 1966

  5. [5]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    V*: Guided visual search as a core mechanism in multimodal llms , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  6. [6]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Argus: Vision-Centric Reasoning with Grounded Chain-of-Thought , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  7. [7]

    The Thirteenth International Conference on Learning Representations , year=

    MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMs , author=. The Thirteenth International Conference on Learning Representations , year=

    work page

  8. [8]

    Chain-of-focus: Adaptive visual search and zooming for multimodal reasoning via rl.arXiv preprint arXiv:2505.15436, 2025

    Chain-of-Focus: Adaptive Visual Search and Zooming for Multimodal Reasoning via RL , author=. arXiv preprint arXiv:2505.15436 , year=

    work page arXiv

  9. [9]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Dyfo: A training-free dynamic focus visual search for enhancing lmms in fine-grained visual understanding , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  10. [10]

    Are multimodal large language models ready for omnidirectional spatial reasoning?, 2025

    Are Multimodal Large Language Models Ready for Omnidirectional Spatial Reasoning? , author=. arXiv preprint arXiv:2505.11907 , year=

    work page arXiv

  11. [11]

    Forty-first International Conference on Machine Learning , year=

    Genie: Generative interactive environments , author=. Forty-first International Conference on Machine Learning , year=

    work page

  12. [12]

    Advances in Neural Information Processing Systems , volume=

    ProcTHOR: Large-Scale Embodied AI Using Procedural Generation , author=. Advances in Neural Information Processing Systems , volume=

    work page

  13. [13]

    2025 , eprint=

    Unfolding Spatial Cognition: Evaluating Multimodal Models on Visual Simulations , author=. 2025 , eprint=

    work page 2025

  14. [14]

    2021 , eprint=

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale , author=. 2021 , eprint=

    work page 2021

  15. [15]

    2025 , eprint=

    Thinking with Images for Multimodal Reasoning: Foundations, Methods, and Future Frontiers , author=. 2025 , eprint=

    work page 2025

  16. [16]

    Chi and F

    Jason Wei and Xuezhi Wang and Dale Schuurmans and Maarten Bosma and Ed H. Chi and F. Xia and Quoc Le and Denny Zhou , booktitle =. ArXiv , title =

    work page

  17. [17]

    Xu and Jun-Mei Song and Mingchuan Zhang and Y

    Zhihong Shao and Peiyi Wang and Qihao Zhu and R. Xu and Jun-Mei Song and Mingchuan Zhang and Y. K. Li and Yu Wu and Daya Guo , booktitle =. ArXiv , title =

    work page

  18. [18]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning , author=. arXiv preprint arXiv:2501.12948 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  19. [19]

    Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models

    Vision-r1: Incentivizing reasoning capability in multimodal large language models , author=. arXiv preprint arXiv:2503.06749 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  20. [20]

    Kimi k1.5: Scaling Reinforcement Learning with LLMs

    Kimi k1. 5: Scaling reinforcement learning with llms , author=. arXiv preprint arXiv:2501.12599 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [21]

    Kimi-VL Technical Report

    Kimi-vl technical report , author=. arXiv preprint arXiv:2504.07491 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [22]

    Kimi K2: Open Agentic Intelligence

    Kimi k2: Open agentic intelligence , author=. arXiv preprint arXiv:2507.20534 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  23. [23]

    2025 , eprint=

    Qwen2.5-VL Technical Report , author=. 2025 , eprint=

    work page 2025

  24. [24]

    2025 , eprint=

    InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency , author=. 2025 , eprint=

    work page 2025

  25. [25]

    NeurIPS , year =

    Liu, Haotian and Li, Chunyuan and Wu, Qingyang and Lee, Yong Jae , title =. NeurIPS , year =

    work page

  26. [26]

    Transactions on Machine Learning Research , year =

    LLaVA-OneVision: Easy Visual Task Transfer , author=. Transactions on Machine Learning Research , year =

    work page

  27. [27]

    International Conference on Machine Learning , pages=

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    work page 2022

  28. [28]

    International Conference on Machine Learning , pages=

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    work page 2023

  29. [29]

    2025 , eprint=

    Gemma 3 Technical Report , author=. 2025 , eprint=

    work page 2025

  30. [30]

    2025 , eprint=

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities , author=. 2025 , eprint=

    work page 2025

  31. [31]

    2025 , month =

    OpenAI , title =. 2025 , month =

    work page 2025

  32. [32]

    2024 , month =

    OpenAI , title =. 2024 , month =

    work page 2024

  33. [33]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year=

    360+x: A Panoptic Multi-modal Scene Understanding Dataset , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year=

    work page

  34. [34]

    2011 , publisher=

    Thinking, fast and slow , author=. 2011 , publisher=

    work page 2011

  35. [35]

    , author=

    Contextual guidance of eye movements and attention in real-world scenes: the role of global features in object search. , author=. Psychological review , volume=. 2006 , publisher=

    work page 2006

  36. [36]

    Proceedings 2003 international conference on image processing (Cat

    Top-down control of visual attention in object detection , author=. Proceedings 2003 international conference on image processing (Cat. No. 03CH37429) , volume=. 2003 , organization=

    work page 2003

  37. [37]

    Advances in neural information processing systems , volume=

    The role of top-down and bottom-up processes in guiding eye movements during visual search , author=. Advances in neural information processing systems , volume=

    work page

  38. [38]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Divide, conquer and combine: A training-free framework for high-resolution image perception in multimodal large language models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  39. [39]

    Mini-o3: Scaling up reasoning patterns and interaction turns for visual search

    Mini-o3: Scaling Up Reasoning Patterns and Interaction Turns for Visual Search , author=. arXiv preprint arXiv:2509.07969 , year=

    work page arXiv

  40. [40]

    Advances in neural information processing systems , volume=

    Flamingo: a visual language model for few-shot learning , author=. Advances in neural information processing systems , volume=

    work page

  41. [41]

    Training language models to follow instructions with human feedback , author=

    work page

  42. [42]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Rlhf-v: Towards trustworthy mllms via behavior alignment from fine-grained correctional human feedback , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  43. [43]

    Advances in Neural Information Processing Systems , volume=

    Cambrian-1: A fully open, vision-centric exploration of multimodal llms , author=. Advances in Neural Information Processing Systems , volume=

    work page

  44. [44]

    Beyond ten turns: Unlocking long-horizon agentic search with large-scale asynchronous rl, 2025

    Beyond ten turns: Unlocking long-horizon agentic search with large-scale asynchronous rl , author=. arXiv preprint arXiv:2508.07976 , year=

    work page arXiv

  45. [45]

    2505.24298 , archivePrefix=

    AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning , author=. arXiv preprint arXiv:2505.24298 , year=

    work page arXiv

  46. [46]

    2025 , url=

    Reinforcing Visual State Reasoning for Multi-Turn VLM Agents , author=. 2025 , url=

    work page 2025

  47. [47]

    2025 , organization =

    OpenManus , title =. 2025 , organization =

    work page 2025

  48. [48]

    Group-in-Group Policy Optimization for LLM Agent Training

    Group-in-Group Policy Optimization for LLM Agent Training , author=. arXiv preprint arXiv:2505.10978 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  49. [49]

    Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning

    Search-r1: Training llms to reason and leverage search engines with reinforcement learning , author=. arXiv preprint arXiv:2503.09516 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  50. [50]

    Royal Society Open Science , volume=

    Eye and head movements are complementary in visual selection , author=. Royal Society Open Science , volume=. 2017 , publisher=

    work page 2017

  51. [51]

    Scientific Reports , volume=

    Eye and head movements in visual search in the extended field of view , author=. Scientific Reports , volume=. 2024 , publisher=

    work page 2024

  52. [52]

    , author=

    Vestibulo-ocular function during co-ordinated head and eye movements to acquire visual targets. , author=. The Journal of Physiology , volume=. 1979 , publisher=

    work page 1979

  53. [53]

    Biological cybernetics , volume=

    Interactions between eye and head control signals can account for movement kinematics , author=. Biological cybernetics , volume=. 2001 , publisher=

    work page 2001

  54. [54]

    Experimental Brain Research , volume=

    Human eye-head coordination in two dimensions under different sensorimotor conditions , author=. Experimental Brain Research , volume=. 1997 , publisher=

    work page 1997

  55. [55]

    Brain research , volume=

    The coordination of eye and head movement during smooth pursuit , author=. Brain research , volume=. 1978 , publisher=

    work page 1978

  56. [56]

    2025 , eprint=

    Kimi-VL Technical Report , author=. 2025 , eprint=

    work page 2025

  57. [57]

    ICLR , year=

    CogCoM: A Visual Language Model with Chain-of-Manipulations Reasoning , author=. ICLR , year=

    work page

  58. [58]

    DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning

    DeepEyes: Incentivizing" Thinking with Images" via Reinforcement Learning , author=. arXiv preprint arXiv:2505.14362 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  59. [59]

    arXiv preprint arXiv:2411.16044 , year=

    Zoomeye: Enhancing multimodal llms with human-like zooming capabilities through tree-based image exploration , author=. arXiv preprint arXiv:2411.16044 , year=

    work page arXiv

  60. [60]

    arXiv preprint arXiv:2403.12966 , year=

    Chain-of-spot: Interactive reasoning improves large vision-language models , author=. arXiv preprint arXiv:2403.12966 , year=

    work page arXiv

  61. [61]

    Cosmos-reason1: From physical common sense to embodied reasoning.arXiv preprint arXiv:2503.15558, 2025

    Cosmos-reason1: From physical common sense to embodied reasoning , author=. arXiv preprint arXiv:2503.15558 , year=

    work page arXiv

  62. [62]

    Gemini Robotics: Bringing AI into the Physical World

    Gemini robotics: Bringing ai into the physical world , author=. arXiv preprint arXiv:2503.20020 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  63. [63]

    Conference on Robot Learning , pages=

    DriveVLM: The Convergence of Autonomous Driving and Large Vision-Language Models , author=. Conference on Robot Learning , pages=. 2025 , organization=

    work page 2025

  64. [64]

    Conference on Robot Learning , pages=

    Robotic Control via Embodied Chain-of-Thought Reasoning , author=. Conference on Robot Learning , pages=. 2025 , organization=

    work page 2025

  65. [65]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Cot-vla: Visual chain-of-thought reasoning for vision-language-action models , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  66. [66]

    Emma: End-to-end multimodal model for autonomous driving.arXiv preprint arXiv:2410.23262, 2024

    Emma: End-to-end multimodal model for autonomous driving , author=. arXiv preprint arXiv:2410.23262 , year=

    work page arXiv

  67. [67]

    Qwen3 Technical Report

    Qwen3 Technical Report , author=. arXiv preprint arXiv:2505.09388 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  68. [68]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) , address=

    LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) , address=. 2024 , url=

    work page 2024

  69. [69]

    Proceedings of The 9th Conference on Robot Learning , pages =

    Eye, Robot: Learning to Look to Act with a BC-RL Perception-Action Loop , author =. Proceedings of The 9th Conference on Robot Learning , pages =. 2025 , editor =

    work page 2025

  70. [70]

    arXiv preprint arXiv:2511.00153 , year=

    EgoMI: Learning Active Vision and Whole-Body Manipulation from Egocentric Human Demonstrations , author=. arXiv preprint arXiv:2511.00153 , year=

    work page arXiv

  71. [71]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Habitat: A platform for embodied ai research , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  72. [72]

    Conference on robot learning , pages=

    CARLA: An open urban driving simulator , author=. Conference on robot learning , pages=. 2017 , organization=

    work page 2017

  73. [73]

    Field and service robotics: Results of the 11th international conference , pages=

    Airsim: High-fidelity visual and physical simulation for autonomous vehicles , author=. Field and service robotics: Results of the 11th international conference , pages=. 2017 , organization=

    work page 2017

  74. [74]

    International Conference on Learning Representation , year=

    MetaUrban: An Embodied AI Simulation Platform for Urban Micromobility , author=. International Conference on Learning Representation , year=

    work page

  75. [75]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Gibson env: Real-world perception for embodied agents , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    work page

  76. [76]

    Conference on Robot Learning , pages=

    Behavior-1k: A benchmark for embodied ai with 1,000 everyday activities and realistic simulation , author=. Conference on Robot Learning , pages=. 2023 , organization=

    work page 2023

  77. [77]

    2017 IEEE international conference on robotics and automation (ICRA) , pages=

    Target-driven visual navigation in indoor scenes using deep reinforcement learning , author=. 2017 IEEE international conference on robotics and automation (ICRA) , pages=. 2017 , organization=

    work page 2017

  78. [78]

    Advances in Neural Information Processing Systems , volume=

    Object goal navigation using goal-oriented semantic exploration , author=. Advances in Neural Information Processing Systems , volume=

    work page

  79. [79]

    Robotics: Science and Systems , year=

    GOAT: GO to Any Thing , author=. Robotics: Science and Systems , year=

    work page

  80. [80]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Neural topological slam for visual navigation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

Showing first 80 references.