Pith. sign in

REVIEW 3 major objections 3 minor 80 references

FUSE: Active Functional Affordance Grounding through Adaptive Semantic-Geometric Evidence Acquisition

T0 review · 3 major / 3 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 FUSE introduces active functional affordance grounding: an entropy-gated planner selects viewpoints from semantic-geometric uncertainty, reaching 72% grounding success at a 1.33x computation saving.

desk verdict The task and benchmark are worth a look, but the 'non-oracle' claim is undercut by the amortized planner's privileged training targets. read the letter →

arxiv 2608.12683 v1 pith:33SJKAZL submitted 2026-08-13 cs.RO cs.CV

Zhou Chen , Sathyanarayanan N. Aakur This is my paper · ORCID
classification cs.ROcs.CV
keywords activefunctionalaffordancegroundinguncertaintyviewpointselectionsemantic-geometricevidenceamortizedplanningexplicitexploration3DGaussianscenerepresentationpartialobservability
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

This paper defines a new embodied perception task: given a functional request such as "something to pour milk in," an agent must move through a partially observed scene until it can identify and spatially locate an object that satisfies the request without being told the object's category in advance. The authors propose FUSE, a framework that decides where to look next by maintaining a spatial map of unresolved semantic and geometric evidence, and that alternates between a fast learned planner and a slower explicit exploration routine depending on how confident the learned planner's action choice is. On a simulated tabletop benchmark of 100 episodes, FUSE succeeds in 72% of episodes with 70.91% mean mask overlap (IoU), the best non-oracle grounding performance reported in the paper, while cutting end-to-end computation by 1.33 times relative to running explicit exploration at every step. The result matters because agents that act on function rather than category need a perception loop that actively resolves ambiguity instead of committing from a single fixed view.

What carries the argument

The load-bearing object is the functional uncertainty field, a viewpoint-conditioned spatial map of where the evidence needed to ground the query's candidate objects is still unresolved. It is the sum of a semantic component, built from per-pixel segmentation confidence for each query-induced candidate label, and a geometric component, defined as the inverse of a structural signal-to-noise ratio between the current observation and the render of an incrementally refined Gaussian 3D scene model. The field becomes an evidence-acquisition map that scores candidate camera actions by the evidence expected in the left, right, upper, or lower image region those actions would reveal. FUSE's adaptive rule is an entropy gate: when the softmax distribution over the amortized planner's predicted action values is below a threshold, the agent follows the fast prediction; otherwise it refines the 3D model and invokes explicit exploration. A grounding-based stopping rule, combining segmentation confidence with depth proximity to the target, ends exploration after K consecutive non-improving views.

What would settle it

Run the benchmark again but judge success by whether the agent can physically perform the queried function with the grounded object (pour, cut, scoop, sip) at the returned viewpoint, instead of by mask IoU; if high-confidence, high-IoU viewpoints do not reliably yield objects that pass the interaction test, the evidence signal is not tracking functional suitability.

Watch

Extended reading notes

Core claim

FUSE's central claim is that task-directed sensing should be driven by a fused functional-uncertainty map instead of generic scene coverage, fixed viewpoints, or a known target category. The map decomposes unresolved evidence into a semantic part, where segmentation confidence for the query's candidate object labels is weak in the current view, and a geometric part, where the observed image disagrees with a render from an incrementally refined Gaussian 3D scene model. Explicit planning scores each candidate camera action by the average acquisition value in the image region that action would expose; a learned amortized planner imitates these scores in a single forward pass; and normalized action entropy decides which of the two to trust. On the introduced benchmark, FUSE reaches 72% grounding success and 70.91% mean IoU, above static grounding (42%), random active exploration (56%), a label-provided VLM active baseline (65%), and either planning mode alone (66% amortized, 70% explicit), while using 1.33 times less computation than full explicit exploration.

Load-bearing premise

The load-bearing premise is that the segmentation model's mask confidence and the overlap of its mask with a ground-truth object mask actually measure whether the found object satisfies the requested function.

Editorial extensions

If this is right

  • Functional grounding no longer needs a predefined target class: the query generates candidate labels that serve as hypotheses, and exploration resolves which hypothesis is grounded.
  • Fixed or predetermined viewpoints are not enough under occlusion, since static and single-canonical-view grounding reach only 42% and 32% success on the benchmark.
  • The entropy gate provides most of explicit search's reliability at lower cost, because FUSE routes about two-thirds of decisions through the amortized branch and still matches or exceeds explicit planning's accuracy.
  • Because gains persist across several affordance knowledge sources, active evidence acquisition and upstream functional-hypothesis generation act as complementary stages rather than interchangeable alternatives.

Reading between the lines

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

  • If FUSE is transferred to a physical robot, the natural next test is to replace mask overlap with physical interaction success; an agent that actually pours, cuts, or scoops with the grounded object would tell whether the semantic-geometric confidence signal tracks function or merely visibility.
  • The benchmark's discrete elliptical camera loop and three pitch levels are restrictive; allowing continuous or six-degree-of-freedom viewpoint control would increase the number of candidate actions and likely raise the value of the amortized planner as a pre-filter.
  • The same confidence-gated fast/deliberative architecture could apply to other active perception problems, such as task-oriented grasping or object search, where a cheap policy should handle routine sensing and expensive model updates should be reserved for ambiguous states.
  • One should not read the 72% figure as an upper bound: the gap to the oracle-label variant (77%) suggests that improved functional hypothesis generation, not only sensing, is a separate route to higher grounding success.
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 / 3 minor

Summary. This paper introduces Active Functional Affordance Grounding, a sequential perception task in which an agent must move a camera to localize an object satisfying a functional query under partial observability. The proposed FUSE framework combines an explicit exploration policy that scores candidate viewpoints from SAM3 semantic evidence and SSNR-based geometric uncertainty (Eqs. 4-5) with a learned amortized planner, switching between the two on the basis of action entropy. The authors contribute a Habitat-based benchmark with 100 fixed episodes, evaluate against passive, random, VLM-driven, explicit, and amortized baselines, and report that FUSE reaches 72% success and 70.91% average IoU (Table 1), a 1.33x end-to-end speedup over Explicit Exploration (Table 5), and robustness across seven affordance knowledge sources (Table 2). The appendix includes paired McNemar tests, bootstrap intervals, ablations of the semantic and geometric components, and detailed runtime profiling.

Significance. The task definition and benchmark are a useful addition to embodied perception: they make explicit the distinction between grounding from a fixed view and actively acquiring functional evidence. The explicit semantic-geometric explorer is a reasonable heuristic, and the entropy-gated integration of amortized and explicit planning is architecturally sensible. The paper is also careful in several ways: all methods are evaluated on the same 100 episode definitions; paired exact McNemar tests and bootstrap confidence intervals are reported; the failure of the success advantage over Explicit Exploration to reach significance is admitted in the appendix; and the cross-knowledge-source comparison is a good stress test. If the central claim of non-oracle performance survives the concerns below, the paper would make a moderate but solid contribution. The main unresolved issues are the privileged training signal used for the amortized planner and the use of the same SAM3 model for both viewpoint selection and grounding evaluation; both are fixable with additional experiments, and neither invalidates the benchmark or the explicit-exploration results.

major comments (3)
  1. [Appendix 'Planner Training Targets'; 'Amortized Evidence Planning' and Table 6] The learned planner is not trained without oracle information. In the appendix, the utility target is y_t = w_sem Δs_t + w_unc Δu_t + w_iou Δq_t + w_vis Δv_t + w_geo Δg_t, where Δq_t is the change in IoU between the highest-confidence SAM3 region and the ground-truth target mask, Δv_t is target visibility normalized by the scene-specific maximum target pixel count, and Δg_t is normalized inverse camera-to-target distance. These are privileged quantities that a non-oracle agent should not have. Since FUSE routes 66.59% of online decisions through the amortized branch (Table 6), and since Success and IoU in 'Metric Computation' are computed by comparing SAM3 masks with the same ground-truth masks, the planner is effectively trained on the evaluation metric. The claim of 'highest observed non-oracle grounding performance' (Abstract; Table 1) is therefore not established: the improvement over Amortized Planning (+6pp, p=0.109) and part of FUSE's gain over Explicit Exploration may reflect supervised fitting to ground-truth target masks rather than an emergent benefit of semantic-geometric evidence acquisition. I ask the authors to retrain the planner using only observable targets, for example changes in SAM3 score and geometric uncertainty without any ground-truth mask, visibility, or distance terms, and to re-report the FUSE comparisons, or to clearly label the current variant as an oracle-assisted bound.
  2. [Eq. (4) and 'Metric Computation'] SAM3 is used both as the semantic evidence source that drives viewpoint selection and as the model whose masks define Success and Average IoU. In Eq. (4), the acquisition map is built from SAM3 confidence and geometric uncertainty; in 'Metric Computation', success is the fraction of episodes in which the highest-scoring SAM3 mask at the returned viewpoint has IoU above 0.5 with a valid ground-truth target. The evaluation therefore rewards finding viewpoints that make SAM3 confident and well-aligned, but it does not independently validate that the grounded object is functionally suitable. All compared methods use SAM3, so the relative comparisons are not unfair, but the absolute claim about reliable functional grounding is weakened. I recommend a control experiment with an independent semantic segmenter at evaluation time, a downstream functional interaction check, or at least an analysis of how the FUSE-vs-baseline differences change when masks for the success metric come from a different model.
  3. [Table 7 and Abstract] The headline '72% success, highest observed non-oracle grounding performance' is a point estimate on a single 100-episode benchmark. The paper's own paired statistics show that FUSE versus Explicit Exploration is not significant (exact McNemar p=0.688, CI [-3,7]pp) and FUSE versus Amortized Planning is not significant at the conventional level (p=0.109, CI [0,12]pp, including zero). The appendix states that 'the current evaluation does not establish a statistically distinguishable success advantage over fully explicit reasoning,' yet the abstract and Table 1 present the point estimate without this qualification. Given the small benchmark and the risk of metric overfitting discussed above, the main claim should be rephrased as comparable grounding performance with a 1.33x computation reduction; additional episodes or a Bayesian analysis would be needed before claiming the highest non-oracle success rate.
minor comments (3)
  1. [Eq. (4) and Appendix 'Semantic and Geometric Evidence'] The form of the evidence map in Eq. (4) does not visibly match the appendix's E_t(u) = w_s S_t(u) + w_g U^geo_t(u) with w_s = w_g = 1 and U^geo scaled by alpha = 0.8; the main text should define the relationship between the gradient-ratio term and the SSNR-based U^geo, and state where alpha enters, to avoid a seeming inconsistency.
  2. [Appendix 'Confidence Gate'] The entropy threshold eta is selected from {0.6, 0.7, 0.8, 0.9} using only five validation episodes, and no sensitivity analysis of eta is reported. Because eta controls the amortized/explicit branch rate and the computational savings, a small sweep of eta on the development set would make the design choice more robust.
  3. [Appendix 'Planner Training Targets'] The amortized planner is trained for a fixed 20 epochs with no checkpoint selection, as the validation loss is monitored but not used for checkpoint selection. Given that this planner handles roughly two-thirds of online decisions, reporting training and validation loss curves or using early stopping would increase confidence that the learned policy has not merely memorized the 416 training trajectories.

Circularity Check

1 steps flagged · score 5.0 of 10

The amortized planner is trained on oracle-defined IoU, visibility, and distance targets that are the same ground-truth quantities used in the evaluation metrics, so the 'non-oracle' success of the learned branch is partly a fitted input relabeled as a prediction; the explicit-exploration branch and knowledge-source robustness keep the rest of the paper empirically independent.

  1. fitted input called prediction [Appendix: Planner Training Targets and Metric Computation; Confidence Gate; Section on Active Functional Grounding results]
    "Here, s_t denotes the maximum SAM3 prompt confidence, u_t denotes the mean 3DGS uncertainty over the nonzero semantic-evidence region, q_t denotes the IoU between the highest-confidence SAM3 region and the ground-truth target mask, v_t denotes target visibility normalized by the scene-specific maximum target pixel count, and g_t denotes normalized inverse camera-to-target distance. ... Across the evaluation set, 66.59% of decisions use the amortized branch and 33.41% use explicit exploration."

    The amortized planner is trained to predict post-action changes in q_t, v_t, and g_t, all of which are defined from the ground-truth target mask and camera-to-target distance. The paper's Success Rate is defined as IoU >= 0.5 between the highest-scoring SAM3 mask and the ground-truth target mask, and Target Pixel Ratio and Distance Score are normalized from ground-truth target pixel counts and camera-target distances. Since 66.59% of online decisions are routed through this planner, the reported non-oracle success is substantially a fit to the evaluation metric rather than an emergent property of adaptive evidence acquisition.

full rationale

The paper's main empirical comparison is self-contained and transparent: all methods share the same fixed episode definitions, SAM3 model, stopping rule, and action budget, and paired bootstrap and McNemar statistics are reported. The explicit exploration branch is a hand-crafted combination of SAM3 semantic confidence and SSNR geometric uncertainty and does not use ground-truth target masks or positions, so the comparisons against Static, Random Active, VLM Active, and the ablations retain independent content. The load-bearing circular step is confined to the amortized planner: its training target y_t includes changes in SAM3-vs-ground-truth IoU, ground-truth-normalized target visibility, and normalized inverse camera-to-target distance, which are exactly the ground-truth quantities used to define Success and viewpoint-quality metrics. Because 66.59% of online decisions are made by this planner, the reported 72% non-oracle success is partly a fitted prediction of the evaluation metric. Nevertheless, the explicit branch alone reaches 70%, the paired FUSE-vs-Explicit difference is not statistically significant, and the knowledge-source robustness study holds the planner fixed while varying upstream reasoners, so the central adaptive-evidence-acquisition framework is not wholly reducible to the fitted target. A score of 5 reflects one substantial fitted-input/called-prediction step without total circularity.

Assumptions & free parameters 9 free parameters · 6 assumptions · 0 invented entities

The framework introduces no new physical entities. The quantified contributions rest on a number of hand-set weights and thresholds, plus assumptions that SAM3 and SSNR are valid evidence signals and that a 100-episode synthetic benchmark supports generalization claims. The entropy threshold is the most consequential fitted choice because it directly sets the amortized/explicit branch rate.

free parameters (9)
  • Entropy threshold eta = 0.8
    Selected by grid search over {0.6,0.7,0.8,0.9} on five validation episodes; controls switching between amortized and explicit planning.
  • Geometric uncertainty scale alpha = 0.8
    Hand-set scale for SSNR-derived geometric uncertainty in all experiments.
  • Evidence fusion weights w_s, w_g = 1.0, 1.0
    Hand-set weights in the evidence-acquisition map E_t.
  • Grounding score weights lambda_s, lambda_d = 0.7, 0.3
    Hand-set weights combining SAM3 confidence and depth proximity for termination.
  • Planner target weights w_sem, w_unc, w_iou, w_vis, w_geo = 1.0, 0.5, 1.0, 0.2, 1.0
    Hand-set weights defining the amortized planner's utility target y_t.
  • Depth proximity normalization range = 0.3-1.2 m
    Hand-set interval for converting depth to a proximity score.
  • Softmax temperature tau = 0.1
    Hand-set temperature for converting amortized action values into the entropy-gated distribution.
  • Patience K = 5
    Number of non-improving observations before termination; protocol choice, ablated against K=3.
  • Sensing budget T = 30
    Maximum online sensing actions per episode; fixed across active methods.
assumptions (6)
  • ad hoc to paper Functional uncertainty can be decomposed additively into semantic and geometric components (Eq. 2)
    The decomposition is a modeling assumption; no derivation shows these components are independent or that their sum captures all relevant evidence.
  • domain assumption SAM3 confidence scores are valid semantic evidence for functional affordance grounding
    SAM3 both drives viewpoint selection and defines the success metric, so the validity of SAM3 as a functional evidence source is load-bearing.
  • domain assumption SSNR gradient discrepancy between rendered and observed images captures task-relevant geometric uncertainty
    Used in Eq. 4 as the geometric exploration signal; no external validation that this quantity is the right uncertainty measure.
  • domain assumption Habitat/iTHOR tabletop scenes are representative of real-world partial observability for functional grounding
    All experiments are in simulation; no real-robot validation is presented, and the authors list it as future work.
  • domain assumption The candidate hypothesis set O_q contains the functionally correct object
    Authors state that active exploration cannot recover a target absent from the hypothesis set, so the method inherits the quality of the upstream knowledge source.
  • domain assumption COLMAP registration and 3DGS refinement yield pose-accurate renderings for SSNR comparison
    The geometric uncertainty signal compares rendered and observed images; alignment errors would corrupt the signal.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FUSE: Active Functional Affordance Grounding through Adaptive Semantic-Geometric Evidence Acquisition." pith.science (2026). https://pith.science/paper/33SJKAZL

@misc{pith2026260812683,
  author       = {Pith},
  title        = {Pith review of: FUSE: Active Functional Affordance Grounding through Adaptive Semantic-Geometric Evidence Acquisition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/33SJKAZL}},
  note         = {Machine review of arXiv:2608.12683}
}
read the original abstract

Embodied agents must often identify and interact with objects based on their function rather than their identity, requiring them to actively acquire observations that reveal discriminative functional evidence. Existing affordance grounding methods operate from fixed viewpoints and lack mechanisms for deciding where to look when functional cues are occluded or incomplete. We introduce Active Functional Affordance Grounding, a new task in which an agent sequentially explores a scene to identify and spatially ground an object satisfying a functional query. To address this problem, we propose FUSE, an adaptive semantic-geometric evidence acquisition framework that combines explicit uncertainty-driven exploration with a learned amortized planner to efficiently select informative viewpoints. We further introduce a Habitat-based benchmark for evaluating active functional grounding. Experiments show that FUSE achieves the highest observed non-oracle grounding performance while reducing computation by 1.33x relative to fully explicit exploration, and remains effective across multiple affordance knowledge sources.

Figures

Figures reproduced from arXiv: 2608.12683 by the authors.

Figure 1
Figure 1. Active functional affordance grounding. Passive grounding predicts from a fixed observation and may commit before sufficient evidence is available. Active grounding ac￾quires additional observations to reduce semantic–geometric uncertainty until a functionally suitable object is grounded. resolved from evidence acquired across viewpoints before physical execution and verification. This formulation places different r… view at source ↗
Figure 2
Figure 2. Overview of FUSE. A functional query is first converted into candidate object hypotheses and grounded using semantic (SAM3) and geometric (SSNR) evidence. FUSE adapts between fast amortized planning and explicit exploration to acquire informative observations until no further improvement is observed, returning the best reliable functional grounding. observations until a suitable target can be reliably grounded. To a… view at source ↗
Figure 3
Figure 3. Visualization of FUSE’s semantic–geometric evidence acquisition. SAM3 highlights candidate functional objects, [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

80 extracted references · 54 canonical work pages

  1. [1]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  2. [2]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  3. [3]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  4. [4]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  5. [5]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    doi:10.1016/s0020-7373(84)80003-6 1984

  6. [6]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  7. [7]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  8. [8]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

Show all 80 references
  1. [9]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  2. [10]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  3. [11]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  4. [12]

    2014 , publisher=

    The ecological approach to visual perception: classic edition , author=. 2014 , publisher=

    2014

  5. [13]

    Proceedings of the IEEE , volume=

    Active perception , author=. Proceedings of the IEEE , volume=. 1988 , publisher=

    1988

  6. [14]

    International journal of computer vision , volume=

    Active vision , author=. International journal of computer vision , volume=. 1988 , publisher=

    1988

  7. [15]

    Proceedings

    The determination of next best views , author=. Proceedings. 1985 IEEE international conference on robotics and automation , volume=. 1985 , organization=

    1985

  8. [16]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Fan, Lei and Zhou, Jianxiong and Xing, Xiaoying and Wu, Ying , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  9. [17]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Fan, Lei and Liang, Mingfu and Li, Yunxuan and Hua, Gang and Wu, Ying , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  10. [18]

    Conference on Robot Learning , pages=

    Affordance-Driven Next-Best-View Planning for Robotic Grasping , author=. Conference on Robot Learning , pages=. 2023 , organization=

    2023

  11. [19]

    2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Closed-loop next-best-view planning for target-driven grasping , author=. 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2022 , organization=

    2022

  12. [20]

    2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    VISO-Grasp: vision-language informed spatial object-centric 6-DoF active view planning and grasping in clutter and invisibility , author=. 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2025 , organization=

    2025

  13. [21]

    arXiv preprint arXiv:2606.19091 , year=

    GCNGrasp-VP: Affordance-Guided View Planning for Efficient Task-Oriented Grasping , author=. arXiv preprint arXiv:2606.19091 , year=

    arXiv

  14. [22]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Lei, Boshu and Jiang, Wen and Daniilidis, Kostas , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2026 , pages =

    2026

  15. [23]

    arXiv preprint arXiv:1505.01085 , year=

    In defense of the direct perception of affordances , author=. arXiv preprint arXiv:1505.01085 , year=

    arXiv

  16. [24]

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

    One-shot open affordance learning with foundation models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  17. [25]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Vuong, An Dinh and Vu, Minh Nhat and Huang, Baoru and Nguyen, Nghia and Le, Hieu and Vo, Thieu and Nguyen, Anh , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  18. [26]

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

    Seqafford: Sequential 3d affordance reasoning via multimodal large language model , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  19. [27]

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

    Open-vocabulary functional 3d scene graphs for real-world indoor spaces , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  20. [28]

    IEEE Robotics and Automation Letters , year=

    EASE: Embodied Active Event Perception via Self-Supervised Energy Minimization , author=. IEEE Robotics and Automation Letters , year=

  21. [29]

    arXiv preprint arXiv:2508.12916 , year=

    RoboRetriever: Single-Camera Robot Object Retrieval via Active and Interactive Perception with Dynamic Scene Graph , author=. arXiv preprint arXiv:2508.12916 , year=

    arXiv

  22. [30]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Liu, Mengzhen and Zhou, Enshen and Chi, Cheng and Han, Yi and Rong, Shanyu and Chen, Liming and Wang, Pengwei and Wang, Zhongyuan and Zhang, Shanghang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2026 , pages =

    2026

  23. [31]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Liu, Zhenyang and Gu, Yongchong and Wang, Yikai and Xue, Xiangyang and Fu, Yanwei , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2026 , pages =

    2026

  24. [32]

    arXiv preprint arXiv:2602.18374 , year=

    Zero-shot Interactive Perception , author=. arXiv preprint arXiv:2602.18374 , year=

  25. [33]

    arXiv preprint arXiv:2204.01691 , year =

    Do As I Can, Not As I Say: Grounding Language in Robotic Affordances , author =. arXiv preprint arXiv:2204.01691 , year =

    arXiv

  26. [34]

    ProgPrompt: Generating Situated Robot Task Plans using Large Language Models , year=

    Singh, Ishika and Blukis, Valts and Mousavian, Arsalan and Goyal, Ankit and Xu, Danfei and Tremblay, Jonathan and Fox, Dieter and Thomason, Jesse and Garg, Animesh , booktitle=. ProgPrompt: Generating Situated Robot Task Plans using Large Language Models , year=

  27. [35]

    Proceedings of The 7th Conference on Robot Learning , pages =

    RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control , author =. Proceedings of The 7th Conference on Robot Learning , pages =. 2023 , volume =

    2023

  28. [36]

    International Conference on Learning Representations (ICLR) , year=

    Vima: General robot manipulation with multimodal prompts , author=. International Conference on Learning Representations (ICLR) , year=

  29. [37]

    Robotics: Science and Systems XIII , year =

    Dex-Net 2.0: Deep Learning to Plan Robust Grasps with Synthetic Point Clouds and Analytic Grasp Metrics , author =. Robotics: Science and Systems XIII , year =

  30. [38]

    IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Fang, Hao-Shu and Wang, Chenxi and Gou, Minghao and Lu, Cewu , title =. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

  31. [39]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Sawatzky, Johann and Souri, Yaser and Grund, Christian and Gall, Jurgen , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

  32. [40]

    2022 , url=

    Pengfei Li and Beiwen Tian and Yongliang Shi and Xiaoxue Chen and Hao Zhao and Guyue Zhou and Ya-Qin Zhang , booktitle=. 2022 , url=

    2022

  33. [41]

    Contact-GraspNet: Efficient 6-DoF Grasp Generation in Cluttered Scenes , year=

    Sundermeyer, Martin and Mousavian, Arsalan and Triebel, Rudolph and Fox, Dieter , booktitle=. Contact-GraspNet: Efficient 6-DoF Grasp Generation in Cluttered Scenes , year=

  34. [42]

    Object Goal Navigation using Goal-Oriented Semantic Exploration , url =

    Chaplot, Devendra Singh and Gandhi, Dhiraj Prakashchand and Gupta, Abhinav and Salakhutdinov, Russ R , booktitle =. Object Goal Navigation using Goal-Oriented Semantic Exploration , url =

  35. [43]

    arXiv preprint arXiv:2512.04231 , year=

    CRAFT-E: A Neuro-Symbolic Framework for Embodied Affordance Grounding , author=. arXiv preprint arXiv:2512.04231 , year=

  36. [44]

    Aakur , booktitle=

    Zhou Chen and Joe Lin and Sathyanarayanan N. Aakur , booktitle=. 2025 , url=

    2025

  37. [45]

    Proceedings of Robotics: Science and Systems , YEAR =

    Thao Nguyen AND Nakul Gopalan AND Roma Patel AND Matthew Corsaro AND Ellie Pavlick AND Stefanie Tellex , TITLE =. Proceedings of Robotics: Science and Systems , YEAR =

  38. [46]

    Conference on Robot Learning , year=

    Same Object, Different Grasps: Data and Semantic Knowledge for Task-Oriented Grasping , author=. Conference on Robot Learning , year=

  39. [47]

    Task-Oriented Grasp Prediction with Visual-Language Inputs , year=

    Tang, Chao and Huang, Dehao and Meng, Lingxiao and Liu, Weiyu and Zhang, Hong , booktitle=. Task-Oriented Grasp Prediction with Visual-Language Inputs , year=

  40. [48]

    AffordGrasp: In-Context Affordance Reasoning for Open-Vocabulary Task-Oriented Grasping in Clutter , year=

    Tang, Yingbo and Zhang, Shuaike and Hao, Xiaoshuai and Wang, Pengwei and Wu, Jianlong and Wang, Zhongyuan and Zhang, Shanghang , booktitle=. AffordGrasp: In-Context Affordance Reasoning for Open-Vocabulary Task-Oriented Grasping in Clutter , year=

  41. [49]

    Open-Vocabulary Functional 3D Scene Graphs for Real-World Indoor Spaces , year=

    Zhang, Chenyangguang and Delitzas, Alexandros and Wang, Fangjinhua and Zhang, Ruida and Ji, Xiangyang and Pollefeys, Marc and Engelmann, Francis , booktitle=. Open-Vocabulary Functional 3D Scene Graphs for Real-World Indoor Spaces , year=

  42. [50]

    , journal=

    Bajcsy, R. , journal=. Active perception , year=

  43. [51]

    International Journal of Computer Vision , year =

    Active Vision , author =. International Journal of Computer Vision , year =

  44. [52]

    and Aakur, Sathyanarayanan N

    Chen, Zhou and Kundu, Sanjoy and Baweja, Harsimran S. and Aakur, Sathyanarayanan N. , journal=. EASE: Embodied Active Event Perception via Self-Supervised Energy Minimization , year=

  45. [53]

    arXiv preprint arXiv:2006.13171 , year=

    Objectnav revisited: On evaluation of embodied agents navigating to objects , author=. arXiv preprint arXiv:2006.13171 , year=

    2006 arXiv

  46. [54]

    arXiv preprint arXiv:2508.05021 , year=

    MAG-Nav: Language-Driven Object Navigation Leveraging Memory-Reserved Active Grounding , author=. arXiv preprint arXiv:2508.05021 , year=

    arXiv

  47. [55]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Chen, Xiao and Li, Quanyi and Wang, Tai and Xue, Tianfan and Pang, Jiangmiao , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  48. [56]

    IEEE Robotics and Automation Letters , year =

    ActiveSplat: High-Fidelity Scene Reconstruction through Active Gaussian Splatting , author =. IEEE Robotics and Automation Letters , year =

  49. [57]

    arXiv , year =

    Matthew Strong and Boshu Lei and Aiden Swann and Wen Jiang and Kostas Daniilidis and Monroe Kennedy III , title =. arXiv , year =

  50. [58]

    Next Best Sense: Guiding Vision and Touch with FisherRF for 3D Gaussian Splatting , year=

    Strong, Matthew and Lei, Boshu and Swann, Aiden and Jiang, Wen and Daniilidis, Kostas and Kennedy, Monroe , booktitle=. Next Best Sense: Guiding Vision and Touch with FisherRF for 3D Gaussian Splatting , year=

  51. [59]

    arXiv preprint arXiv:2511.04199 , year=

    GraspView: Active Perception Scoring and Best-View Optimization for Robotic Grasping in Cluttered Environments , author=. arXiv preprint arXiv:2511.04199 , year=

  52. [60]

    Proceedings of the 36th Annual Conference of the Cognitive Science Society , year =

    Amortized Inference in Probabilistic Reasoning , author =. Proceedings of the 36th Annual Conference of the Cognitive Science Society , year =

  53. [61]

    Advances in Variational Inference , year=

    Zhang, Cheng and Bütepage, Judith and Kjellström, Hedvig and Mandt, Stephan , journal=. Advances in Variational Inference , year=

  54. [62]

    Journal of Artificial Intelligence Research , volume=

    Amortized variational inference: A systematic review , author=. Journal of Artificial Intelligence Research , volume=

  55. [63]

    Entropy-Gated Branching for Efficient Test-Time Reasoning

    Li, Xianzhi and Callanan, Ethan and Ghassel, Abdellah and Zhu, Xiaodan. Entropy-Gated Branching for Efficient Test-Time Reasoning. Proceedings of the 19th Conference of the E uropean Chapter of the A ssociation for C omputational L inguistics (Volume 1: Long Papers). 2026. doi...

    2026 doi

  56. [64]

    arXiv preprint arXiv:2601.04442 , year=

    Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning Optimization , author=. arXiv preprint arXiv:2601.04442 , year=

    arXiv

  57. [65]

    Think or Not? Selective Reasoning via Reinforcement Learning for Vision-Language Models , url =

    Wang, Jiaqi and Lin, Kevin Qinghong and Cheng, James and Shou, Mike Zheng , booktitle =. Think or Not? Selective Reasoning via Reinforcement Learning for Vision-Language Models , url =

  58. [66]

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

    Self-supervised learning from images with a joint-embedding predictive architecture , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  59. [67]

    ACM Transactions On Graphics (TOG) , volume=

    A hierarchical 3d gaussian representation for real-time rendering of very large datasets , author=. ACM Transactions On Graphics (TOG) , volume=. 2024 , publisher=

    2024

  60. [68]

    2018 , copyright =

    Ha, David and Schmidhuber, Jürgen , title =. 2018 , copyright =. doi:10.5281/ZENODO.1207631 , url =

    2018 doi

  61. [69]

    Robotics and Autonomous Systems , volume=

    Active object recognition by view integration and reinforcement learning , author=. Robotics and Autonomous Systems , volume=. 2000 , publisher=

    2000

  62. [70]

    arXiv preprint arXiv:2511.16719 , year=

    Sam 3: Segment anything with concepts , author=. arXiv preprint arXiv:2511.16719 , year=

    arXiv

  63. [71]

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

    What object should i use?-task driven object detection , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  64. [72]

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

    Structure-from-motion revisited , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  65. [73]

    3D Gaussian Splatting for Real-Time Radiance Field Rendering , journal =

    Kerbl, Bernhard and Kopanas, Georgios and Leimk. 3D Gaussian Splatting for Real-Time Radiance Field Rendering , journal =. 2023 , url =

    2023

  66. [74]

    arXiv preprint arXiv:2507.06261 , year=

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities , author=. arXiv preprint arXiv:2507.06261 , year=

    arXiv

  67. [75]

    arXiv preprint arXiv:2407.21783 , year=

    The llama 3 herd of models , author=. arXiv preprint arXiv:2407.21783 , year=

    arXiv

  68. [76]

    2025 , institution =

    GPT-5 System Card , author =. 2025 , institution =

    2025

  69. [77]

    2026 , url =

    Claude 3.6 Sonnet Model , author =. 2026 , url =

    2026

  70. [78]

    2023 , archivePrefix=

    Habitat 3.0: A Co-Habitat for Humans, Avatars and Robots , author =. 2023 , archivePrefix=

    2023

  71. [79]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Habitat 2.0: Training Home Assistants to Rearrange their Habitat , author =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  72. [80]

    Habitat:

    Manolis Savva and Abhishek Kadian and Oleksandr Maksymets and Yili Zhao and Erik Wijmans and Bhavana Jain and Julian Straub and Jia Liu and Vladlen Koltun and Jitendra Malik and Devi Parikh and Dhruv Batra , booktitle =. Habitat:

Pith tools

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