Pith. sign in

REVIEW 4 major objections 7 minor 1 cited by

Beyond Transformers: Linear Attention Policy for Open-Vocabulary Object Goal Navigation

T0 review · 4 major / 7 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read This paper claims that linear-attention policy backbones, which compress history into structured recurrent states, outperform RNN and Transformer backbones for open-vocabulary object goal navigation, with a proposed WSLA variant reaching 36

desk verdict A worthwhile empirical study of linear-attention backbones for OVON with credible three-seed results, but the headline comparison is partly confounded by tying rollout length to context length and an undertuned Transformer baseline. read the letter →

arxiv 2607.18794 v1 pith:6KNMFNXH submitted 2026-07-21 cs.RO

Jiahong Zhang , Yifan Lin , Yandong Zhang , Sijun Shen , Kexin Wang , Yuqi Pan , Hongjuan Pei , Wei Wang
show 1 more author
Guoqi Li
This is my paper
classification cs.RO
keywords open-vocabularyobjectgoalnavigationlinearattentionpolicybackbonestate-spacemodelspartiallyobservablesuccessrateembodiedWSLA
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper is trying to establish that the way a navigation policy updates its internal state—not how much history it can attend to—is the key driver of performance under partial observability. It argues that self-attention-based Transformers, despite their explicit context window, do not exploit longer histories on the open-vocabulary object goal navigation (OVON) task. In contrast, linear-attention backbones, which maintain a structured recurrent state updated from each observation, consistently outperform both RNN and Transformer baselines under matched settings. The authors introduce WSLA, a linear-attention variant that expands each head into multiple regulated sub-states with learned readout weighting, achieving a 6.3 percentage-point average success-rate gain over the best Transformer. If true, this points to a concrete design direction for embodied policies: invest in state-update design rather than context-length scaling.

What carries the argument

The central object is the linear-attention head as a recurrent associative memory: instead of retaining a window of tokens and recomputing self-attention, each head maintains a state matrix S_t that is incrementally updated from the current key-value pair and read out by the current query. WSLA extends each of the H heads into E independent sub-states, each with its own retention decay and write coefficient (a gated delta rule), and then aggregates the sub-state readouts with a learnable softmax-weighted combination shared across heads and time. This machinery replaces quadratic-window self-attention with a near-constant per-step update that compresses history into a fixed-size structured st

What would settle it

Train the Transformer baseline with a per-backbone hyperparameter search (learning rate, rollout length, context length) or with the same state-update inductive bias; if it then matches or exceeds the 36.4% average success rate on HM3D-OVON, the claim that self-attention cannot exploit longer histories in OVON fails.

Watch

Extended reading notes

Core claim

Under a controlled comparison that fixes visual and language encoders, training pipeline, and optimization settings, the paper finds that linear-attention policy backbones outperform both RNN and Transformer backbones on HM3D-OVON. Transformer policies do not improve with longer training contexts—performance can even degrade—whereas linear-attention backbones improve with larger training context lengths and stay stable across test-time lengths. The proposed WSLA, which expands each attention head into multiple sub-states with independent gated delta updates and a learnable weighted readout, reaches a 36.4% average success rate versus 30.1% for the strongest Transformer baseline, with the lar

Load-bearing premise

The comparison assumes the imitation-and-PPO training recipe, with rollout length matched to context length, suits the Transformer as well as the linear-attention backbones; if the Transformer baseline were simply undertuned, the reported gap would shrink.

Editorial extensions

If this is right

  • Linear-attention backbones provide a third design choice beyond RNNs and Transformers for partially observable navigation policies, improving average success rate by 6.3 points over the best Transformer baseline on HM3D-OVON.
  • Transformer policies do not benefit from longer training contexts in OVON, whereas linear-attention backbones do, suggesting the state-update mechanism is more important than the context window for history integration.
  • Within the linear-attention family, more structured and regulated state-update rules (Linear Transformer → DeltaNet → Gated DeltaNet → WSLA) yield progressively better navigation performance.
  • WSLA transfers to the HSSD benchmark with a 9.6-point zero-shot SR gain and an 8.3-point fine-tuned SR gain over the Transformer, and it increases SPL substantially after fine-tuning, indicating more efficient adaptation.
  • WSLA maintains near-constant online FLOPs (about 0.048G per action) as context length grows, unlike the Transformer whose cost grows quadratically, making the approach more practical for on-device or long-horizon deployment.

Reading between the lines

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

  • The same state-expansion-plus-weighted-readout recipe could transfer to other partially observable RL tasks beyond navigation—such as manipulation or search—where the policy must decide what to write into memory and when to overwrite it.
  • The paper's finding that Transformer gains do not scale with context length under the given training recipe suggests that matching rollout length to context window may not be the right training protocol for Transformer policies; a per-backbone schedule could narrow the gap.
  • Since the largest gains appear on long-distance episodes, a testable extension would be to make the state-update rule explicitly sensitive to goal-relevant cues, e.g., by learning a write gate conditioned on the semantic similarity between the current observation and the goal description.
  • The constant per-step cost of linear-attention backbones suggests they could extend beyond the 500-step training horizon used here, and the paper's bounded-context limitation could be relaxed by allowing the state to persist across episode resets.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes LANav, a linear-attention policy backbone for Open-Vocabulary Object Goal Navigation (OVON), and its variant WSLA, which expands each attention head into multiple sub-states with learned weighted readout. The central claim is that, under matched training and architecture settings, linear-attention/state-update backbones outperform RNN and Transformer baselines on HM3D-OVON, with WSLA reaching 36.4% average SR versus 30.1% for the best Transformer, and that the advantage grows with longer training contexts. The paper also reports distance-stratified gains, HSSD zero-shot and fine-tuned transfer, and a real-robot deployment on a Unitree Go2 with 82% success over 50 trials.

Significance. If the central claim holds, the paper identifies a concrete and potentially general design principle for partially observable embodied policies: structured recurrent state updates are more effective than fixed-window self-attention for integrating navigation history, independent of sequence-length complexity. The study is significant in scope because it uses an external benchmark (HM3D-OVON), matches the policy backbone while fixing encoders and training pipeline, reports three-seed means and standard deviations for the main table, and includes both simulated and real-world transfer results. The parameter-count and FLOPs analysis in Fig. 4 also makes the efficiency argument concrete. However, the claim is load-bearing on the fairness of the shared training recipe, and several supporting analyses lack uncertainty quantification or a decoupled experimental design.

major comments (4)
  1. [Sec. 4.1 / Fig. 2] The central 'LA benefits from longer contexts; Transformer does not' claim is confounded. Training Details states 'the rollout length is set to match the context length l.' Figure 2 varies l_train in {100, 300, 500}, thus simultaneously varying the PPO rollout length. Longer rollouts alter advantage estimation, value-target distribution, and update frequency, which can independently affect Transformer optimization. The reported plateau/degradation of Transformer at l_train=500 may therefore be an artifact of the rollout schedule, not a property of self-attention. Please rerun Fig. 2 with rollout length fixed while varying only the context window, or otherwise decouple these two variables.
  2. [Table 1 / Sec. 4.1] All reproduced methods share Adam lr=2.5e-4, no warmup, and the DAgRL recipe. Transformers are commonly sensitive to learning rate and warmup, and the reproduced Transformer at 30.1% may be undertuned relative to the LA variants. Since the headline 6.3 pp gap between WSLA and Transformer is the paper's main evidence, a learning-rate or warmup sensitivity study for the Transformer baseline is required to show the comparison is not an optimization mismatch. At minimum, report results with a Transformer tuned under the same budget (e.g., lr sweep or a brief warmup schedule).
  3. [Fig. 2 / Table 4] Figure 2 and the HSSD results (Table 4) are reported without error bars or multiple seeds. The main table includes three-seed means, but the context-length and HSSD claims are equally central. Please provide means and standard deviations over at least three seeds for these analyses, or clearly state if only one seed was used and justify why the observed differences are significant.
  4. [Method / Related Work] The paper positions LA as an alternative to both RNN and Transformer, and argues that structured state update is the key mechanism. However, no selective state-space model (e.g., S4, Mamba, or a gated SSM baseline) is included in the comparisons. Such models also maintain a structured recurrent state and are widely used in decision-making. Without this baseline, the claim that linear attention specifically—rather than any recurrent state-update backbone—is responsible for the gains remains incomplete. Please add at least one selective-SSM baseline under matched settings, or explicitly delimit the claim to the LA family.
minor comments (7)
  1. [Abstract / Page 1] The abstract uses 'constant FLOPs' phrasing in the efficiency claim; Fig. 4 shows near-constant but not exactly constant online FLOPs. Suggest 'nearly constant' for precision.
  2. [Sec. 3.3 / Eq. (18)] The RMSNorm applied after concatenation is not described in the text; please specify whether it is applied over the feature dimension only, and whether the gate input z_t is the backbone-layer hidden state.
  3. [Fig. 4] The right panel uses a log-scale y-axis without labeling it as log. Add a clear axis label so the reader can interpret the FLOPs comparison directly.
  4. [Sec. 4.2 / Table 1] The EALM row is marked N/A for VAL SEEN SYN and is described as using a different training strategy. Clarify in the table caption whether EALM uses the same visual and language encoders, since that affects comparability.
  5. [Sec. 4.4 / Table 3] The ablation in Table 3 reports SR increases of 34.6->35.2->36.4 without error bars. Given the main table shows seed-level variation of ~0.3, please report means and standard deviations for the ablation as well.
  6. [Conclusion] The limitations sentence is brief; it mentions shallow backbones and bounded context windows but not the absence of code. Since the paper states that code will be made public, please either provide a link in the final version or explicitly state that it is available upon request.
  7. [General] Several references to the authors' own prior work (MetaLA, Spiking-Brain, MMDend) appear in the related work and LA family discussion. While these citations are topically relevant, please double-check that they are cited for their technical content and not simply as self-promotion; the current text appears acceptable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is an external-benchmark comparison, not a derivation that reduces to its inputs.

full rationale

The paper's central claim is an empirical comparison of policy backbones on an external benchmark (HM3D-OVON) with fixed visual/text encoders, training pipeline, and optimization settings. The headline result (WSLA 36.4% SR vs Transformer 30.1%) is not derived from an equation that embeds the claim; it is measured. No parameter is fitted to the reported success-rate numbers and then renamed a prediction; the HM3D-OVON, HSSD transfer, and real-robot results are external evaluations. Self-citations (Chou et al. 2024; Pan et al. 2025a,b; Wang et al. 2025; Chen et al. 2025) appear in related work and in the statement that WSLA is 'inspired by' prior state-expansion studies; none is invoked as a proof of superiority, a uniqueness theorem, or an ansatz that fixes the conclusion. The only identified concern—that 'rollout length is set to match the context length' (Sec. 4.1) confounds the Fig. 2 scaling comparison—is a training-recipe fairness issue, not circularity: it does not make any prediction equal to an input by construction. Accordingly, no circular step is found.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim is empirical, not a first-principles derivation. The ledger lists the manually chosen hyperparameters and domain assumptions that the comparison rests on. No physical entities are postulated; WSLA sub-states are internal architectural constructs rather than new physical objects.

free parameters (5)
  • Context length l=500 = 500
    Selected for all reproduced attention models; context-scaling experiments show WSLA performance changes with l, so the central comparison depends on this choice.
  • Hidden dimension d = 512
    Backbone width used for all models; parameter counts and FLOPs of WSLA depend on it.
  • Heads H and expansion factor E = H=4, E=4
    WSLA capacity hyperparameters; ablation only varies presence/absence of expansion and weighting, not the values of H and E.
  • Training budget = 150M DAgger + 100M PPO
    All models compared at the same fixed budget; relative rankings could change at different training scales.
  • Optimizer hyperparameters = Adam lr=2.5e-4, eps=1e-5, grad clip 0.2
    Shared across backbones; a Transformer or LA variant could behave differently with per-backbone tuning.
assumptions (5)
  • domain assumption OVON is a POMDP whose success is measured by stopping within a threshold of an instance of the goal category
    Inherited from HM3D-OVON; used to define the policy objective and evaluation.
  • domain assumption Frozen SigLIP visual and text encoders provide sufficiently rich representations for all compared backbones
    All models share the same encoders; if encoder quality is the bottleneck, the backbone comparisons may not generalize to other encoders.
  • domain assumption The DAgRL DAgger-then-PPO training recipe is a fair protocol for every backbone
    The recipe and hyperparameters are taken from the LSTM-based DAgRL pipeline and applied unchanged to Transformer/LA models.
  • standard math The linear-attention update equations in Supplementary S1 faithfully instantiate Linear Transformer, DeltaNet, and Gated DeltaNet
    The paper relies on cited prior formulations; any implementation discrepancy would affect the comparison.
  • domain assumption The discrete action set and velocity adapter used on the Unitree Go2 preserve the semantics of the learned policy's action space
    Real-robot deployment drops look-up/look-down and maps forward/turn commands to velocities; if this mapping shifts the action distribution, sim-to-real conclusions weaken.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Transformers: Linear Attention Policy for Open-Vocabulary Object Goal Navigation." pith.science (2026). https://pith.science/paper/6KNMFNXH

@misc{pith2026260718794,
  author       = {Pith},
  title        = {Pith review of: Beyond Transformers: Linear Attention Policy for Open-Vocabulary Object Goal Navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6KNMFNXH}},
  note         = {Machine review of arXiv:2607.18794}
}
read the original abstract

Open-Vocabulary Object Goal Navigation (OVON) requires agents to operate under partial observability, making effective internal state updates critical for navigation performance. This update is implemented by the policy network, where recent approaches adopt Transformer-based backbones with self-attention over a context window to integrate temporal information. However, our controlled experiments show that performance does not scale with context length under Transformer-based policies, questioning the suitability of self-attention for state integration in navigation. To this end, we propose Linear Attention-based Navigation (LANav), which adopts linear attention (LA) as the policy backbone to maintain a structured state update rather than self-attention over the context window. Across multiple LA variants evaluated under identical settings, LANav consistently outperforms Transformer-based baselines. Performance improves as state update mechanisms become more structured and regulated, highlighting the importance of state update design. To improve state update effectiveness, we introduce Weighted State-Expansion Linear Attention (WSLA), which expands each attention head's state into multiple sub-states and uses learnable weighted readout to aggregate expanded sub-states. Equipped with WSLA, LANav achieves 36.4% average success rate (SR) on HM3D-OVON, outperforming Transformer-based counterparts by 6.3 percentage points in macro-averaged SR, while maintaining computational efficiency. Distance-stratified results show larger gains in long-distance episodes, while HSSD transfer and fine-tuning demonstrate robustness across scene distributions. Real-world deployment on a Unitree Go2 further achieves an 82% success rate over 50 trials, supporting the practical feasibility and sim-to-real transfer of LANav.

Figures

Figures reproduced from arXiv: 2607.18794 by the authors.

Figure 1
Figure 1. Overview of LANav. Left: Open-Vocabulary Object Goal Navigation task. Middle: At step t, the RGB observation It, goal description G, and previous action at−1 are encoded by a vision encoder, a text encoder, and an action embedding module, respectively. The resulting features form a per-step embedding o˜t. A context window consisting of the most recent embeddings {o˜t−(l−1), . . . , o˜t} is processed by the policy ba… view at source ↗
Figure 2
Figure 2. Effect of context length. Overall SR (averaged across validation splits) as a function of test-time context length ltest. Each curve corresponds to a model trained with ltrain ∈ {100, 300, 500}. goal names. Within the LA family, performance improves from Linear Transformer to DeltaNet, Gated DeltaNet, and finally WSLA, suggesting that more structured and regulated state updates lead to better navigation performance.… view at source ↗
Figure 3
Figure 3. Qualitative comparison of navigation behaviors on HM3D-OVON. Trans LT GDN WSLA Model 0 5 10 15 20 25 Parameters (M) 100 300 500 1000 Context length 10 1 10 0 10 1 Online FLOPs / action (G) Transformer (Trans) Linear Transformer (LT) Gated DeltaNet (GDN) WSLA (Ours) [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Parameter count and computational cost. Model parameters (left) and online FLOPs (right) of different policy backbones under identical architectural settings. Analysis of WSLA We conduct ablation studies to ana￾lyze the effects of state expansion and weighted aggregati…

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. SpikingNav: Robust Embodied Navigation with Spiking Neural Policies

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A spiking sensing encoder and spiking policy network improve ObjectNav success under visual corruptions (8.45% to 13.71%) while using fewer parameters and fewer FLOPs than a matched ANN baseline.

Reference graph

Works this paper leans on

59 extracted references · 4 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

    Savva, Manolis and Kadian, Abhishek and Maksymets, Oleksandr and Zhao, Yili and Wijmans, Erik and Jain, Bhavana and Straub, Julian and Liu, Jia and Koltun, Vladlen and Malik, Jitendra and others , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

  2. [2]

    Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing , pages=

    Learning phrase representations using RNN encoder--decoder for statistical machine translation , author=. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing , pages=

    2014

  3. [3]

    Neural Computation , volume=

    Long short-term memory , author=. Neural Computation , volume=

  4. [4]

    arXiv preprint arXiv:1807.06757 , year=

    On evaluation of embodied navigation agents , author=. arXiv preprint arXiv:1807.06757 , year=

    arXiv

  5. [5]

    Neural Networks , volume=

    Sigmoid-weighted linear units for neural network function approximation in reinforcement learning , author=. Neural Networks , volume=. 2018 , publisher=

    2018

  6. [6]

    arXiv preprint arXiv:2507.16577 , year=

    Scaling Linear Attention with Sparse State Expansion , author=. arXiv preprint arXiv:2507.16577 , year=

    arXiv

  7. [7]

    Advances in Neural Information Processing Systems , volume=

    Root mean square layer normalization , author=. Advances in Neural Information Processing Systems , volume=

  8. [8]

    Auxiliary tasks and exploration enable

    Ye, Joel and Batra, Dhruv and Das, Abhishek and Wijmans, Erik , booktitle=. Auxiliary tasks and exploration enable

Show all 59 references
  1. [9]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Scaling up the state size of RNN LLMs for long-context scenarios , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  2. [10]

    arXiv preprint arXiv:1707.06347 , year=

    Proximal policy optimization algorithms , author=. arXiv preprint arXiv:1707.06347 , year=

    arXiv

  3. [11]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Sigmoid loss for language image pre-training , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  4. [12]

    International Conference on Machine Learning , pages=

    Linear transformers are secretly fast weight programmers , author=. International Conference on Machine Learning , pages=

  5. [13]

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

    Scene memory transformer for embodied agents in long-horizon tasks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  6. [14]

    Advances in Neural Information Processing Systems , volume=

    Generalization of reinforcement learners with working and episodic memory , author=. Advances in Neural Information Processing Systems , volume=

  7. [15]

    Advances in Neural Information Processing Systems , volume=

    Attention is all you need , author=. Advances in Neural Information Processing Systems , volume=

  8. [16]

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

    Visual navigation with spatial attention , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  9. [17]

    2022 International Conference on Robotics and Automation , pages=

    Object memory transformer for object goal navigation , author=. 2022 International Conference on Robotics and Automation , pages=

    2022

  10. [18]

    IEEE Robotics and Automation Letters , volume=

    Transformer memory for interactive visual navigation in cluttered environments , author=. IEEE Robotics and Automation Letters , volume=

  11. [19]

    2025 , eprint=

    OVSegDT: Segmenting Transformer for Open-Vocabulary Object Goal Navigation , author=. 2025 , eprint=

    2025

  12. [20]

    Transformers are

    Katharopoulos, Angelos and Vyas, Apoorv and Pappas, Nikolaos and Fleuret, Fran. Transformers are. International Conference on Machine Learning , pages=

  13. [21]

    Advances in Neural Information Processing Systems , volume=

    Metala: Unified optimal linear approximation to softmax attention map , author=. Advances in Neural Information Processing Systems , volume=

  14. [22]

    International Conference on Machine Learning , year=

    Gated Linear Attention Transformers with Hardware-Efficient Training , author=. International Conference on Machine Learning , year=

  15. [23]

    arXiv preprint arXiv:2509.05276 , year=

    SpikingBrain: Spiking Brain-inspired Large Models , author=. arXiv preprint arXiv:2509.05276 , year=

    arXiv

  16. [24]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Mmdend: Dendrite-inspired multi-branch multi-compartment parallel spiking neuron for sequence modeling , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  17. [25]

    Advances in Neural Information Processing Systems , volume=

    Parallelizing linear transformers with the delta rule over sequence length , author=. Advances in Neural Information Processing Systems , volume=

  18. [26]

    and He, Zhiyang and Sax, Alexander and Malik, Jitendra and Savarese, Silvio , title =

    Xia, Fei and Zamir, Amir R. and He, Zhiyang and Sax, Alexander and Malik, Jitendra and Savarese, Silvio , title =. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , year =

  19. [27]

    2017 International Conference on 3D Vision , year =

    Chang, Angel and Dai, Angela and Funkhouser, Thomas and Halber, Maciej and Niessner, Matthias and Savva, Manolis and Song, Shuran and Zeng, Andy and Zhang, Yinda , title =. 2017 International Conference on 3D Vision , year =

    2017

  20. [28]

    Image and Vision Computing , volume=

    Memory-MambaNav: Enhancing object-goal navigation through integration of spatial--temporal scanning with state space models , author=. Image and Vision Computing , volume=. 2025 , publisher=

    2025

  21. [29]

    International Conference on Learning Representations , year=

    Learning to Navigate in Complex Environments , author=. International Conference on Learning Representations , year=

  22. [30]

    International Conference on Learning Representations , year=

    DD-PPO: Learning Near-Perfect PointGoal Navigators from 2.5 Billion Frames , author=. International Conference on Learning Representations , year=

  23. [31]

    Workshop on Reincarnating Reinforcement Learning at ICLR 2023 , year=

    Offline visual representation learning for embodied navigation , author=. Workshop on Reincarnating Reinforcement Learning at ICLR 2023 , year=

    2023

  24. [32]

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

    Habitat-web: Learning embodied object-search strategies from human demonstrations at scale , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  25. [33]

    Advances in Neural Information Processing Systems , year =

    Chaplot, Devendra Singh and Gandhi, Dhiraj and Gupta, Abhinav and Gupta, Saurabh and Salakhutdinov, Ruslan , title =. Advances in Neural Information Processing Systems , year =

  26. [34]

    Advances in Neural Information Processing Systems , year =

    Yin, Hang and Xu, Xiuwei and Wu, Zhenyu and Zhou, Jie and Lu, Jiwen , title =. Advances in Neural Information Processing Systems , year =

  27. [35]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

    Cao, Yihan and Zhang, Jiazhao and Yu, Zhinan and Liu, Shuzhen and Qin, Zheng and Zou, Qin and Du, Bo and Xu, Kai , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , year =

  28. [36]

    arXiv preprint arXiv:2307.08621 , year=

    Retentive network: A successor to transformer for large language models , author=. arXiv preprint arXiv:2307.08621 , year=

    arXiv

  29. [37]

    International Conference on Learning Representations , year=

    Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer , author=. International Conference on Learning Representations , year=

  30. [38]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Deepseekmoe: Towards ultimate expert specialization in mixture-of-experts language models , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  31. [39]

    International Conference on Learning Representations , year =

    Gated Delta Networks: Improving Mamba2 with Delta Rule , author=. International Conference on Learning Representations , year =

  32. [40]

    Advances in Neural Information Processing Systems , year =

    Chen, Lili and Lu, Kevin and Rajeswaran, Aravind and Lee, Aditya and Grover, Aviral and Abbeel, Pieter and Gupta, Arjun and Levine, Sergey , title =. Advances in Neural Information Processing Systems , year =

  33. [41]

    Decision

    David, Shmuel Bar and Zimerman, Itamar and Nachmani, Eliya and Wolf, Lior , booktitle=. Decision

  34. [42]

    Advances in Neural Information Processing Systems , volume=

    Structured state space models for in-context reinforcement learning , author=. Advances in Neural Information Processing Systems , volume=

  35. [43]

    Advances in Neural Information Processing Systems , volume=

    Is mamba compatible with trajectory optimization in offline reinforcement learning? , author=. Advances in Neural Information Processing Systems , volume=

  36. [44]

    Advances in Neural Information Processing Systems , year =

    Ota, Kentaro and others , title =. Advances in Neural Information Processing Systems , year =

  37. [45]

    Ramrakhya, Ram and Batra, Dhruv and Wijmans, Erik and Das, Abhishek , booktitle=

  38. [46]

    2024 IEEE International Conference on Robotics and Automation , pages=

    Exploitation-guided exploration for semantic embodied navigation , author=. 2024 IEEE International Conference on Robotics and Automation , pages=

    2024

  39. [47]

    Conference on Robot Learning , pages=

    PoliFormer: Scaling On-Policy RL with Transformers Results in Masterful Navigators , author=. Conference on Robot Learning , pages=

  40. [48]

    2023 IEEE/RSJ International Conference on Intelligent Robots and Systems , pages=

    Control transformer: Robot navigation in unknown environments through prm-guided return-conditioned sequence modeling , author=. 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems , pages=

    2023

  41. [49]

    IEEE Robotics and Automation Letters , volume=

    NavFormer: A transformer architecture for robot target-driven navigation in unknown and dynamic environments , author=. IEEE Robotics and Automation Letters , volume=

  42. [50]

    Proceedings of the 7th ACM International Conference on Multimedia in Asia , articleno =

    Chen, Zifan and Zhang, Jiahong and Shen, Sijun and Fu, Wenlong and Cao, Lihong , title =. Proceedings of the 7th ACM International Conference on Multimedia in Asia , articleno =. 2025 , isbn =

    2025

  43. [51]

    HM3D-OVON: A Dataset and Benchmark for Open-Vocabulary Object Goal Navigation , year=

    Yokoyama, Naoki and Ramrakhya, Ram and Das, Abhishek and Batra, Dhruv and Ha, Sehoon , booktitle=. HM3D-OVON: A Dataset and Benchmark for Open-Vocabulary Object Goal Navigation , year=

  44. [52]

    2022 , eprint=

    VER: Scaling On-Policy RL Leads to the Emergence of Navigation in Embodied Rearrangement , author=. 2022 , eprint=

    2022

  45. [53]

    2011 , eprint=

    A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning , author=. 2011 , eprint=

    2011

  46. [54]

    2023 , eprint=

    LLaMA: Open and Efficient Foundation Language Models , author=. 2023 , eprint=

    2023

  47. [55]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

    Move to Understand a 3D Scene: Bridging Visual Grounding and Exploration for Efficient and Versatile Embodied Navigation , author =. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

  48. [56]

    Ziliotto, Filippo and Campari, Tommaso and Serafini, Luciano and Ballan, Lamberto , booktitle =

  49. [57]

    Zemskova, Tatiana and Staroverov, Aleksei and Yudin, Dmitry and Panov, Aleksandr , journal =

  50. [58]

    Yokoyama, Naoki and Ha, Sehoon and Batra, Dhruv and Wang, Jiuguang and Bucher, Bernadette , booktitle =

  51. [59]

    and Savva, Manolis , title =

    Khanna, Mukul and Mao, Yongsen and Jiang, Hanxiao and Haresh, Sanjay and Shacklett, Brennan and Batra, Dhruv and Clegg, Alexander and Undersander, Eric and Chang, Angel X. and Savva, Manolis , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog...

    2024

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.