Pith. sign in

REVIEW 3 major objections 2 minor 54 references

A discounted Beta-Bernoulli estimator turns noisy group rewards into stable advantages, making verifiable RL for LLM reasoning far more sample-efficient.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-13 22:35 UTC pith:SUQX6L7C

load-bearing objection We only have the abstract for DBB; the supplied full text is a different paper (SR-Nav), so the theory and Acc@8 claims cannot be checked. the 3 major comments →

arxiv 2603.18444 v2 pith:SUQX6L7C submitted 2026-03-19 cs.LG cs.AI

Discounted Beta-Bernoulli Reward Estimation for Sample-Efficient Reinforcement Learning with Verifiable Rewards

Haechan Kim , Soohyun Ryu , Gyouk Chu , Doohyuk Jang , Eunho Yang This is my paper
classification cs.LG cs.AI
keywords reinforcement learning with verifiable rewardssample efficiencyBeta-Bernoulli estimationadvantage estimationGRPOLLM reasoningvariance collapse
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Group-based reinforcement learning with verifiable rewards (RLVR) for large language models wastes most of its generated responses because it treats a handful of rollouts as a point estimate of reward, producing high-variance or collapsed advantages. This paper recasts the problem as statistical estimation of a policy-induced reward distribution and introduces Discounted Beta-Bernoulli (DBB) estimation, which folds historical success/failure counts into a discounted Beta posterior. The resulting estimator is biased yet has lower and more stable variance, theoretically avoids estimated-variance collapse, and yields lower mean-squared error than the usual point estimate. When plugged into GRPO it raises Acc@8 by several points on in-distribution reasoning tasks and by larger margins out-of-distribution, all without extra compute or memory. A reader who cares about post-training reasoning models therefore gains a drop-in way to extract more learning signal from the same number of rollouts.

Core claim

Modeling binary verifiable rewards as draws from a non-stationary Bernoulli whose success probability is induced by the current policy, then estimating that probability with a discounted Beta-Bernoulli posterior that re-uses historical counts, produces advantage estimates whose variance stays controlled, never collapses to zero, and has lower MSE than the empirical mean of a small group of rollouts; substituting these advantages into GRPO therefore improves sample efficiency and final accuracy on reasoning benchmarks.

What carries the argument

Discounted Beta-Bernoulli (DBB) reward estimation: a biased posterior mean that mixes the current group’s binary rewards with exponentially discounted historical success/failure counts, thereby regularizing the reward distribution estimate used for advantage computation.

Load-bearing premise

Historical reward counts, even after discounting, remain informative enough about the current policy’s reward distribution that the resulting biased estimator still has lower mean-squared error than an unbiased point estimate of a tiny group.

What would settle it

Run the identical GRPO training loop on the same model and data with and without DBB; if the Acc@8 gap on the six in-distribution and three out-of-distribution reasoning suites disappears (or reverses) while the empirical variance of the advantage estimates does not collapse under the naïve estimator, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Existing group-based RLVR pipelines can replace their point-estimate advantage step with DBB and obtain higher final accuracy for free.
  • Advantage estimates remain usable even when a group is all-correct or all-incorrect, eliminating the variance-collapse failure mode that wastes whole batches.
  • Out-of-distribution reasoning gains are larger than in-distribution gains, suggesting the estimator improves generalization of the learned policy.
  • The same statistical view can be applied to other binary or low-cardinality verifiable reward signals beyond the GRPO setting.

Where Pith is reading between the lines

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

  • Because the method only needs running success/failure counts, it can be layered on top of any online RL algorithm that already stores recent rewards, not just GRPO.
  • If the discount factor is annealed more aggressively as the policy stabilizes, the bias–variance trade-off might improve further late in training.
  • The same Beta-Bernoulli construction could stabilize multi-objective or preference-based RL where reward signals are similarly sparse and binary.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The abstract claims that group-based RLVR (e.g., GRPO) is sample-inefficient because it uses high-variance point estimates of binary rewards from few rollouts, which can collapse estimated variance and waste generated responses. The authors reframe advantage estimation as finite-sample estimation of a policy-induced reward distribution and propose Discounted Beta-Bernoulli (DBB) estimation, which folds discounted historical reward statistics into a Beta-Bernoulli posterior. They assert that the resulting biased estimator has lower and more stable variance, theoretically avoids estimated variance collapse, and attains lower MSE than point estimation, and that GRPO+DBB yields average Acc@8 gains of 3.22/2.42 (in-distribution) and 12.49/6.92 (out-of-distribution) on 1.7B/8B models with no extra compute or memory. The supplied full manuscript body, however, is an unrelated paper (SR-Nav on zero-shot object-goal navigation), so the DBB theory, algorithms, and experimental tables cannot be inspected.

Significance. If the abstract claims hold, the work would be a practically useful and theoretically motivated fix for a known pain point in RLVR post-training of reasoning LLMs: cheap, drop-in reward estimation that improves both ID and especially OOD Acc@8 without extra rollouts, memory, or compute. A clean bias–variance analysis showing that a discounted Beta-Bernoulli posterior avoids variance collapse under non-stationary policy-induced rewards would also be a reusable statistical contribution for group-based RL methods. Those strengths cannot be credited from the materials provided, because the full text does not contain the DBB derivation, proofs, or GRPO experiments.

major comments (3)
  1. Manuscript mismatch: the CACHEABLE full text is SR-Nav (object-goal navigation; arXiv-style header 2603.18443), not the DBB/RLVR paper named in the abstract and paper_id 2603.18444. No DBB estimator, no Beta-Bernoulli update, no GRPO advantage formula, and no Acc@8 tables appear in the body. The central claims are therefore unverifiable from the submitted materials.
  2. Abstract theory claim (bias, stable variance, avoidance of estimated variance collapse, lower MSE vs point estimation) cannot be checked: there is no equation defining the discounted historical sufficient statistics, no statement of the non-stationary reward model, and no theorem or proof sketch. The load-bearing assumption that discounted history remains informative under a changing policy is unsupported in the provided text.
  3. Abstract experimental claim (Acc@8 +3.22/+2.42 ID and +12.49/+6.92 OOD on 1.7B/8B; six ID and three OOD benchmarks; no extra cost) cannot be audited: no tables, seeds, group sizes, baselines beyond “naive GRPO,” ablations of the discount/prior, or error bars are present. Reported gains must be treated as unconfirmed.
minor comments (2)
  1. Even at abstract level, free parameters (historical discount, Beta prior hyperparameters, group size) are not specified; a camera-ready version should state defaults and sensitivity.
  2. Clarify whether “verifiable rewards” are strictly binary (0/1) so that Beta-Bernoulli is conjugate, or whether multi-valued rewards are reduced to Bernoulli.

Circularity Check

0 steps flagged

No circularity identifiable: abstract states a biased Beta-Bernoulli estimator with claimed lower MSE and external Acc@8 gains; supplied full text is a different paper (SR-Nav), so no load-bearing derivation can be reduced to its inputs.

full rationale

The target paper (Discounted Beta-Bernoulli / GRPO+DBB) is available only as an abstract. That abstract frames advantage estimation as finite-sample estimation of a policy-induced reward distribution, proposes a discounted Beta-Bernoulli posterior that reuses historical binary rewards, and asserts reduced/stable variance, avoidance of estimated variance collapse, lower MSE than point estimation, and Acc@8 gains on external reasoning benchmarks. None of these claims, as stated, is definitionally equal to a fitted free parameter, a self-defined quantity, or a uniqueness theorem imported from the same authors. There is no equation chain, no self-citation of a prior uniqueness result, and no renaming of a known empirical pattern that can be exhibited as circular from the given text. The CACHEABLE full manuscript is SR-Nav (object-goal navigation with a Dynamic Spatial Relationship Graph), which is a different work and contains no Beta-Bernoulli, GRPO, or RLVR derivation to inspect. Per the hard rules, circularity may be claimed only when a specific reduction can be quoted; that is impossible here. Residual risk that discount/prior hyperparameters were tuned to the same eval suites is a correctness/tuning concern, not circularity. Score 0 with empty steps is the honest outcome.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 1 invented entities

Abstract-only review of 2603.18444 (full text in context is a different paper). Load-bearing structure is: binary verifiable rewards ~ Bernoulli under the current policy; conjugate Beta prior plus discounted historical counts yield a usable estimate under non-stationarity; lower MSE / no variance collapse of that estimator improves GRPO. Free parameters (discount, Beta hyperparameters, group size) are implied but not numerically specified. No new physical entities; the invented object is the DBB estimator itself.

free parameters (3)
  • historical discount factor (for past reward statistics)
    Abstract says DBB 'leverages historical reward statistics for the non-stationary distribution' via discounting; the discount rate is a free design choice that controls bias-variance and is not fixed by theory in the abstract.
  • Beta prior hyperparameters (alpha, beta)
    Beta-Bernoulli estimation requires prior pseudo-counts; abstract does not specify defaults or how they are set, yet they shape the estimator's bias and early-training behavior.
  • number of rollouts / group size for GRPO
    Point-estimation baseline and DBB both depend on finite group size; sample-efficiency claims are relative to that choice, which is standard but free.
axioms (3)
  • domain assumption Verifiable rewards are binary (correct/incorrect) and can be modeled as i.i.d. Bernoulli draws from a policy-induced success probability for a given prompt.
    Abstract: 'modeling rewards as samples drawn from a policy-induced distribution' and Beta-Bernoulli conjugacy; standard for RLVR but not true for graded or noisy rewards.
  • ad hoc to paper Despite non-stationarity of the policy (and thus of the reward distribution), discounted historical reward statistics remain informative enough that a biased DBB estimator has lower MSE than current-batch point estimates and avoids estimated variance collapse.
    Core modeling bet of the paper; abstract asserts it theoretically but provides no proof text here.
  • domain assumption Advantage estimates with lower MSE / non-collapsed variance improve GRPO training and final Acc@8 on reasoning benchmarks.
    Links statistical estimation quality to RL optimization outcomes; common in policy-gradient practice but not automatic.
invented entities (1)
  • Discounted Beta-Bernoulli (DBB) reward estimator no independent evidence
    purpose: Estimate the reward distribution / advantages for group RLVR using current rollouts plus discounted historical reward counts under a Beta-Bernoulli model.
    Named method introduced in the abstract; classical conjugate update plus discounting, packaged for GRPO-style RLVR.

pith-pipeline@v1.1.0-grok45 · 22959 in / 3195 out tokens · 34286 ms · 2026-07-13T22:35:50.896716+00:00 · methodology

0 comments
read the original abstract

Reinforcement learning with verifiable rewards (RLVR) has emerged as an effective post-training paradigm for improving the reasoning capabilities of large language models. However, existing group-based RLVR methods often suffer from severe sample inefficiency. This inefficiency stems from reliance on point estimation of rewards from a small number of rollouts, leading to high estimation variance, variance collapse, and ineffective utilization of generated responses. In this work, we reformulate RLVR from a statistical estimation perspective by modeling rewards as samples drawn from a policy-induced distribution and casting advantage computation as the problem of estimating the reward distribution from finite data. Building on this view, we propose Discounted Beta-Bernoulli (DBB) reward estimation, which leverages historical reward statistics for the non-stationary distribution. Although biased, the resulting estimator exhibits reduced and stable variance, theoretically avoids estimated variance collapse, and achieves lower mean squared error than standard point estimation. Extensive experiments across six in-distribution and three out-of-distribution reasoning benchmarks demonstrate that GRPO with DBB consistently outperforms naive GRPO, achieving average Acc@8 improvements of 3.22/2.42 points in-distribution and 12.49/6.92 points out-of-distribution on the 1.7B and 8B models, respectively, without additional computational cost or memory usage.

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

54 extracted references · 7 linked inside Pith

  1. [1]

    Objectnav revisited: On evaluation of embodied agents navigating to objects,

    D. Batra, A. Gokaslan, A. Kembhavi, O. Maksymets, R. Mottaghi, M. Savva, A. Toshev, and E. Wijmans, “Objectnav revisited: On evaluation of embodied agents navigating to objects,”arXiv preprint arXiv:2006.13171, 2020

    Pith/arXiv arXiv 2006

  2. [2]

    Bridging zero-shot object navigation and foundation models through pixel-guided navigation skill,

    W. Cai, S. Huang, G. Cheng, Y . Long, P. Gao, C. Sun, and H. Dong, “Bridging zero-shot object navigation and foundation models through pixel-guided navigation skill,” inProc. IEEE Int. Conf. Robot. Autom., 2024, pp. 5228–5234

    2024

  3. [3]

    V oila: Visual- observation-only imitation learning for autonomous navigation,

    H. Karnan, G. Warnell, X. Xiao, and P. Stone, “V oila: Visual- observation-only imitation learning for autonomous navigation,” inProc. IEEE Int. Conf. Robot. Autom., 2022, pp. 2497–2503

    2022

  4. [4]

    Zson: Zero-shot object-goal navigation using multimodal goal embed- dings,

    A. Majumdar, G. Aggarwal, B. Devnani, J. Hoffman, and D. Batra, “Zson: Zero-shot object-goal navigation using multimodal goal embed- dings,” inAdv. Neural Inform. Process. Syst., vol. 35, 2022, pp. 32 340– 32 352

    2022

  5. [5]

    Zero-shot active visual search (zavis): Intelligent object search for robotic assistants,

    J. Park, T. Yoon, J. Hong, Y . Yu, M. Pan, and S. Choi, “Zero-shot active visual search (zavis): Intelligent object search for robotic assistants,” in Proc. IEEE Int. Conf. Robot. Autom., 2023, pp. 2004–2010

    2023

  6. [6]

    Zero-shot object goal visual navigation,

    Q. Zhao, L. Zhang, B. He, H. Qiao, and Z. Liu, “Zero-shot object goal visual navigation,” inProc. IEEE Int. Conf. Robot. Autom., 2023, pp. 2025–2031

    2023

  7. [7]

    Hierarchies of planning and reinforcement learning for robot navigation,

    J. W ¨ohlke, F. Schmitt, and H. van Hoof, “Hierarchies of planning and reinforcement learning for robot navigation,” inProc. IEEE Int. Conf. Robot. Autom., 2021, pp. 10 682–10 688

    2021

  8. [8]

    Esc: Exploration with soft commonsense constraints for zero- shot object navigation,

    K. Zhou, K. Zheng, C. Pryor, Y . Shen, H. Jin, L. Getoor, and X. E. Wang, “Esc: Exploration with soft commonsense constraints for zero- shot object navigation,” inProc. Int. Conf. Mach. Learn., 2023, pp. 42 829–42 842

    2023

  9. [9]

    Sg-nav: Online 3d scene graph prompting for llm-based zero-shot object navigation,

    H. Yin, X. Xu, Z. Wu, J. Zhou, and J. Lu, “Sg-nav: Online 3d scene graph prompting for llm-based zero-shot object navigation,” inAdv. Neural Inform. Process. Syst., vol. 37, 2024, pp. 5285–5307

    2024

  10. [10]

    Vlfm: Vision- language frontier maps for zero-shot semantic navigation,

    N. Yokoyama, S. Ha, D. Batra, J. Wang, and B. Bucher, “Vlfm: Vision- language frontier maps for zero-shot semantic navigation,” inProc. IEEE Int. Conf. Robot. Autom., 2024, pp. 42–48

    2024

  11. [11]

    V oronav: voronoi-based zero-shot object navigation with large language model,

    P. Wu, Y . Mu, B. Wu, Y . Hou, J. Ma, S. Zhang, and C. Liu, “V oronav: voronoi-based zero-shot object navigation with large language model,” inProc. Int. Conf. Mach. Learn., 2024, pp. 53 737 – 53 775

    2024

  12. [12]

    Topv-nav: Unlocking the top-view spatial reasoning potential of mllm for zero-shot object navigation,

    L. Zhong, C. Gao, Z. Ding, Y . Liao, H. Ma, S. Zhang, X. Zhou, and S. Liu, “Topv-nav: Unlocking the top-view spatial reasoning potential of mllm for zero-shot object navigation,”arXiv preprint arXiv:2411.16425, 2025

    Pith/arXiv arXiv 2025

  13. [13]

    Cows on pasture: Baselines and benchmarks for language-driven zero-shot object navigation,

    S. Y . Gadre, M. Wortsman, G. Ilharco, L. Schmidt, and S. Song, “Cows on pasture: Baselines and benchmarks for language-driven zero-shot object navigation,” inIEEE Conf. Comput. Vis. Pattern Recog., 2023, pp. 23 171–23 181

    2023

  14. [14]

    Star-searcher: A complete and efficient aerial system for autonomous target search in complex unknown environments,

    Y . Luo, Z. Zhuang, N. Pan, C. Feng, S. Shen, F. Gao, H. Cheng, and B. Zhou, “Star-searcher: A complete and efficient aerial system for autonomous target search in complex unknown environments,”IEEE Robot. Autom. Lett., no. 5, p. 4329–4336, May 2024

    2024

  15. [15]

    Occupancy grids: A stochastic spatial representation for active robot perception,

    A. Elfes, “Occupancy grids: A stochastic spatial representation for active robot perception,”arXiv preprint arXiv, vol. abs/1304.1098, 2013

    Pith/arXiv arXiv 2013

  16. [16]

    A frontier-based approach for autonomous exploration,

    B. Yamauchi, “A frontier-based approach for autonomous exploration,” inProc. IEEE Int. Conf. Robot. Autom., 1997, pp. 146–151

    1997

  17. [17]

    Information gain-based exploration using rao-blackwellized particle filters,

    C. Stachniss, G. Grisetti, and W. Burgard, “Information gain-based exploration using rao-blackwellized particle filters,” inProc. Robot.: Sci. Syst., 06 2005, pp. 65–72

    2005

  18. [18]

    Gamap: Zero-shot object goal navigation with multi-scale geometric-affordance guidance,

    H. Huang, Y . Hao, C. Wen, A. Tzes, Y . Fanget al., “Gamap: Zero-shot object goal navigation with multi-scale geometric-affordance guidance,” inAdv. Neural Inform. Process. Syst., vol. 37, 2024, pp. 39 386–39 408

    2024

  19. [19]

    L3mvn: Leveraging large language models for visual target navigation,

    B. Yu, H. Kasaei, and M. Cao, “L3mvn: Leveraging large language models for visual target navigation,” inProc. IEEE/RSJ Int. Conf. Intell. Robots Syst., 2023, pp. 3554–3560

    2023

  20. [20]

    Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models,

    J. Li, D. Li, S. Savarese, and S. Hoi, “Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models,” inProc. Int. Conf. Mach. Learn., 2023, pp. 19 730–19 742

    2023

  21. [21]

    Qwen2. 5-vl technical report,

    S. Bai, K. Chen, X. Liu, J. Wang, W. Ge, S. Song, K. Dang, P. Wang et al., “Qwen2. 5-vl technical report,”arXiv preprint arXiv:2502.13923, 2025

    Pith/arXiv arXiv 2025

  22. [22]

    Gpt-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidtet al., “Gpt-4 technical report,”arXiv preprint arXiv:2303.08774, 2023

    Pith/arXiv arXiv 2023

  23. [23]

    OpenFMNav: Towards open-set zero-shot object navigation via vision-language foundation models,

    Y . Kuang, H. Lin, and M. Jiang, “OpenFMNav: Towards open-set zero-shot object navigation via vision-language foundation models,” in Proc. Annu. Meet. Assoc. Comput. Linguist., K. Duh, H. Gomez, and S. Bethard, Eds., Jun. 2024, pp. 338–351

    2024

  24. [24]

    Egtr: Extracting graph from transformer for scene graph generation,

    J. Im, J. Nam, N. Park, H. Lee, and S. Park, “Egtr: Extracting graph from transformer for scene graph generation,” inIEEE Conf. Comput. Vis. Pattern Recog., June 2024, pp. 24 229–24 238

    2024

  25. [25]

    Hi- erarchical Open-V ocabulary 3D Scene Graphs for Language-Grounded Robot Navigation,

    A. Werby, C. Huang, M. B ¨uchner, A. Valada, and W. Burgard, “Hi- erarchical Open-V ocabulary 3D Scene Graphs for Language-Grounded Robot Navigation,” inProc. Robot.: Sci. Syst., Delft, Netherlands, July 2024

    2024

  26. [26]

    Expanding scene graph boundaries: Fully open-vocabulary scene graph generation via visual- concept alignment and retention,

    Z. Chen, J. Wu, Z. Lei, Z. Zhang, and C. Chen, “Expanding scene graph boundaries: Fully open-vocabulary scene graph generation via visual- concept alignment and retention,” inEur. Conf. Comput. Vis., 2024, pp. 108–124

    2024

  27. [27]

    Cognav: Cognitive process modeling for object goal navigation with llms,

    Y . Cao, J. Zhang, Z. Yu, S. Liu, Z. Qin, Q. Zou, B. Du, and K. Xu, “Cognav: Cognitive process modeling for object goal navigation with llms,” inInt. Conf. Comput. Vis., 2025, pp. 9550–9560

    2025

  28. [28]

    Beliefmapnav: 3d voxel- based belief map for zero-shot object navigation,

    Z. Zhou, Y . Hu, L. Zhang, Z. Li, and S. Chen, “Beliefmapnav: 3d voxel- based belief map for zero-shot object navigation,” inAdv. Neural Inform. Process. Syst., 2025

    2025

  29. [29]

    Object goal navigation using goal-oriented semantic exploration,

    D. S. Chaplot, D. P. Gandhi, A. Gupta, and R. R. Salakhutdinov, “Object goal navigation using goal-oriented semantic exploration,”Adv. Neural Inform. Process. Syst., vol. 33, pp. 4247–4258, 2020. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13

    2020

  30. [30]

    Habitat-matterport 3d dataset (HM3D): 1000 large-scale 3d environments for embodied AI,

    S. K. Ramakrishnan, A. Gokaslan, E. Wijmans, O. Maksymets, A. Clegg, J. M. Turner, E. Undersander, W. Galuba, A. Westbury, A. X. Chang, M. Savva, Y . Zhao, and D. Batra, “Habitat-matterport 3d dataset (HM3D): 1000 large-scale 3d environments for embodied AI,”arXiv preprint arXiv:2109.08238, 2021

    Pith/arXiv arXiv 2021

  31. [31]

    Matterport3d: Learning from rgb-d data in indoor environments,

    A. Chang, A. Dai, T. Funkhouser, M. Halber, M. Niessner, M. Savva, S. Song, A. Zeng, and Y . Zhang, “Matterport3d: Learning from rgb-d data in indoor environments,”Proc. Int. Conf. 3D Vis., p. 667–676, 2017

    2017

  32. [32]

    Trihelper: Zero-shot object navigation with dynamic assistance,

    L. Zhang, Q. Zhang, H. Wang, E. Xiao, Z. Jiang, H. Chen, and R. Xu, “Trihelper: Zero-shot object navigation with dynamic assistance,” in Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst., 2024, pp. 10 035–10 042

    2024

  33. [33]

    How to not train your dragon: Training-free embodied object goal navigation with semantic frontiers,

    J. Chen, G. Li, S. Kumar, B. Ghanem, and F. Yu, “How to not train your dragon: Training-free embodied object goal navigation with semantic frontiers,” inProc. Robot.: Sci. Syst., 07 2023

    2023

  34. [34]

    Fuel: Fast uav exploration using incremental frontier structure and hierarchical planning,

    B. Zhou, Y . Zhang, X. Chen, and S. Shen, “Fuel: Fast uav exploration using incremental frontier structure and hierarchical planning,”IEEE Robot. Autom. Lett., vol. 6, no. 2, pp. 779–786, 2021

    2021

  35. [35]

    Think holistically, act down-to-earth: A semantic navigation strategy with continuous environmental representation and multi-step forward planning,

    B. Chen, J. Kang, P. Zhong, Y . Cui, S. Lu, Y . Liang, and J. Wang, “Think holistically, act down-to-earth: A semantic navigation strategy with continuous environmental representation and multi-step forward planning,”IEEE Trans. Circuit Syst. Video Technol., vol. 34, no. 5, pp. 3860–3875, 2024

    2024

  36. [36]

    Open scene graphs for open-world object- goal navigation,

    J. Loo, Z. Wu, and D. Hsu, “Open scene graphs for open-world object- goal navigation,” inProc. IEEE Int. Conf. Robot. Autom., 2024

    2024

  37. [37]

    Agent-centric relation graph for object visual navigation,

    X. S. Hu, Z. Wu, K. Lv, S. Wang, and Y . Lin, “Agent-centric relation graph for object visual navigation,”IEEE Trans. Circuit Syst. Video Technol., vol. 34, pp. 1295–1309, 2021

    2021

  38. [38]

    Imaginenav: Prompting vision- language models as embodied navigator through scene imagination,

    X. Zhao, W. Cai, L. Tang, and T. Wang, “Imaginenav: Prompting vision- language models as embodied navigator through scene imagination,” in Int. Conf. Learn. Represent., 2025

    2025

  39. [39]

    BERT: Pre- training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre- training of deep bidirectional transformers for language understanding,” inProc. Annu. Meet. Assoc. Comput. Linguist., J. Burstein, C. Doran, and T. Solorio, Eds., Jun. 2019, pp. 4171–4186

    2019

  40. [40]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,

    D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Maet al., “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,”Nature., p. 633–638, 2025

    2025

  41. [41]

    Chatnav: Leveraging llm to zero-shot semantic reasoning in object navigation,

    Y . Zhu, Z. Wen, X. Li, X. Shi, X. Wu, H. Dong, and J. Chen, “Chatnav: Leveraging llm to zero-shot semantic reasoning in object navigation,” IEEE Trans. Circuit Syst. Video Technol., vol. 35, no. 3, pp. 2369–2381, 2025

    2025

  42. [42]

    E²ba: Environment exploration and backtracking agent for visual language object navigation,

    Y . Shi, J. Liu, L. Sun, and X. Zheng, “E²ba: Environment exploration and backtracking agent for visual language object navigation,”IEEE Trans. Circuit Syst. Video Technol., vol. 35, no. 7, pp. 6231–6244, 2025

    2025

  43. [43]

    Strive: Structured representation integrating vlm reasoning for efficient object navigation,

    H. Zhu, Z. Li, Z. Liu, W. Wang, J. Zhang, J. Francis, and J. Oh, “Strive: Structured representation integrating vlm reasoning for efficient object navigation,” inProc. IEEE Int. Conf. Robot. Autom., 2026

    2026

  44. [44]

    Visual semantic navigation using scene priors,

    W. Yang, X. Wang, A. Farhadi, A. Gupta, and R. Mottaghi, “Visual semantic navigation using scene priors,” inInt. Conf. Learn. Represent., May 2019

    2019

  45. [45]

    Imagine before go: Self-supervised generative map for object goal navigation,

    S. Zhang, X. Yu, X. Song, X. Wang, and S. Jiang, “Imagine before go: Self-supervised generative map for object goal navigation,” inIEEE Conf. Comput. Vis. Pattern Recog., 2024, pp. 16 414–16 425

    2024

  46. [46]

    Hierarchical object-to-zone graph for object navigation,

    S. Zhang, X. Song, Y . Bai, W. Li, Y . Chu, and S. Jiang, “Hierarchical object-to-zone graph for object navigation,” inInt. Conf. Comput. Vis., October 2021, pp. 15 130–15 140

    2021

  47. [47]

    Search for or navigate to? dual adaptive thinking for object navigation,

    R. Dang, L. Wang, Z. He, S. Su, J. Tang, C. Liu, and Q. Chen, “Search for or navigate to? dual adaptive thinking for object navigation,” inInt. Conf. Comput. Vis., 2023, pp. 8216–8225

    2023

  48. [48]

    Layout-based causal inference for object navigation,

    S. Zhang, X. Song, W. Li, Y . Bai, X. Yu, and S. Jiang, “Layout-based causal inference for object navigation,” inIEEE Conf. Comput. Vis. Pattern Recog., 2023, pp. 10 792–10 802

    2023

  49. [49]

    Poni: Potential functions for objectgoal navigation with interaction-free learning,

    S. K. Ramakrishnan, D. S. Chaplot, Z. Al-Halah, J. Malik, and K. Grauman, “Poni: Potential functions for objectgoal navigation with interaction-free learning,” inIEEE Conf. Comput. Vis. Pattern Recog., 2022, pp. 18 890–18 900

    2022

  50. [50]

    Learning object relation graph and tentative policy for visual navigation,

    H. Du, X. Yu, and L. Zheng, “Learning object relation graph and tentative policy for visual navigation,” inEur. Conf. Comput. Vis.Berlin, Heidelberg: Springer-Verlag, 2020, p. 19–34

    2020

  51. [51]

    Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors,

    C.-Y . Wang, A. Bochkovskiy, and H.-Y . M. Liao, “Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors,” inIEEE Conf. Comput. Vis. Pattern Recog., 2023, pp. 7464–7475

    2023

  52. [52]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection,

    S. Liu, Z. Zeng, T. Ren, F. Li, H. Zhang, J. Yang, Q. Jiang, C. Liet al., “Grounding dino: Marrying dino with grounded pre-training for open-set object detection,” inEur. Conf. Comput. Vis., 2024, pp. 38–55

    2024

  53. [53]

    Faster segment anything: Towards lightweight sam for mobile applications,

    C. Zhang, D. Han, Y . Qiao, J. U. Kim, S.-H. Bae, S. Lee, and C. S. Hong, “Faster segment anything: Towards lightweight sam for mobile applications,”arXiv preprint arXiv:2306.14289, 2023

    Pith/arXiv arXiv 2023

  54. [54]

    Unigoal: Towards universal zero-shot goal-oriented navigation,

    H. Yin, X. Xu, L. Zhao, Z. Wang, J. Zhou, and J. Lu, “Unigoal: Towards universal zero-shot goal-oriented navigation,” inIEEE Conf. Comput. Vis. Pattern Recog., 2025, pp. 19 057–19 066

    2025