DreamFly: Causal Memory and Receding-Horizon Diffusion Planning for Aerial Vision-Language Navigation

Fei Xu; Yan Deng

REVIEW 2 major objections 4 minor 31 references

DreamFly: Causal Memory and Receding-Horizon Diffusion Planning for Aerial Vision-Language Navigation

T0 review · 2 major / 4 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 DreamFly shows that causal memory, K-step diffusion planning, and explicit stopping together improve aerial vision-language navigation, with 32.04% seen and 29.46% unseen success on OpenFly.

desk verdict Worth a close read for the memory and stop-head design, but the SOTA claim does not survive contact with the OpenFly-Agent baseline setup. read the letter →

arxiv 2608.12308 v1 pith:TA5IDNNQ submitted 2026-08-12 cs.CV cs.AI

Yan Deng , Fei Xu This is my paper · ORCID

classification cs.CVcs.AI

keywords aerialvision-languagenavigationvision-language-actionmodelsdiffusionpolicylong-termvisualmemoryreceding-horizonplanningactionchunkingexplicitterminationOpenFlybenchmark

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 sets out to establish that three temporal mechanisms, handled together, materially improve instruction-guided aerial navigation: a causally aligned memory that admits only observations from before the current step, a receding-horizon diffusion planner that predicts a K-step action chunk but executes only one action before replanning, and a separate stop head that predicts termination from the policy's initial all-mask logits. The authors argue that most prior aerial VLN systems conflate or omit one of these roles, leaving them unable to retain landmarks, exploit short-horizon lookahead, and avoid overshooting the goal. On the OpenFly benchmark, DreamFly reports success rates of 32.04% and 29.46% and SPL of 28.22% and 23.54% on test-seen and test-unseen splits, the best among the compared baselines, with the lowest navigation error. The central claim is that the gains come from the joint design: each component fixes a distinct temporal failure mode, and ablations removing any one component degrade performance.

What carries the argument

The mechanism that carries the argument is a closed-loop cycle: observe, retrieve from causal memory, denoise an action chunk, assess termination, execute one action, then update memory. The causal boundary is enforced by M_<t = F_mem(I, (O_tau)_tau<t), with a read-before-write protocol so the current observation cannot influence its own decision; the memory is a 16-slot table built by instruction-conditioned candidate extraction from frozen visual routers, where each slot holds anchor and prototype features plus a last-update timestamp, and retrieval is masked cross-attention with a learned gate. The planning half reuses the bidirectional diffusion backbone of Dream-VLA to denoise K masked action tokens with valid-prefix supervision and horizon-dependent loss weights. Termination is separated out: LiteStop maps the initial all-mask action logit grid to a stop probability, and the final decision ORs LiteStop with the frozen policy's own Stop action. The design's working parts are the fixed temporal boundary, the plan-K execute-one loop, and the decoupled stop head.

What would settle it

Run DreamFly closed-loop on the test-unseen split twice: once with its normal online memory, and once where each step's memory is replaced by the expert-prefix memory for the same trajectory. If SR and SPL remain at roughly 29.46% and 23.54%, the causal-memory result is robust; if they drop toward the reported no-memory ablation, the gains depend on training-time memory statistics rather than on the online memory mechanism itself.

Watch

Extended reading notes

Core claim

The central discovery is that all three temporal decisions can be made explicit inside one discrete diffusion policy without changing the action space. The memory branch maintains a fixed set of 16 long-term slots, written with a read-before-write rule so the current observation is never available to the historical branch at its own decision step; a gated cross-attention adapter injects the retrieved context into current visual tokens. The planner denoises K=4 action tokens in a single bidirectional pass, trains only on the valid prefix with horizon decay gamma=0.7, and at runtime executes the leading action, discards the rest, and replans from the next observation. LiteStop consumes the K by |A| logit grid from the initial all-mask forward and is trained separately on binary stop labels with positive-class weight 4.0, so it can stop before the motion policy acts. In closed loop this yields 32.04%/29.46% success rate and 28.22%/23.54% SPL on OpenFly seen/unseen splits, and the ablations show each component contributes.

Load-bearing premise

At deployment the agent must make decisions from memories built from its own online rollouts, while training used memories built from expert trajectories; the paper states these memory states may differ and does not measure or correct that gap, so if online memory states fall outside the training distribution the closed-loop gains could shrink.

Editorial extensions

If this is right

Because future chunk positions are supervised only as auxiliary targets and never executed, the policy can be evaluated at any replanning cadence without changing the action space or the stop mechanism.
The read-before-write memory boundary means any improvement from memory cannot be attributed to peeking at current or future observations, strengthening the causal interpretation of the ablation results.
Training LiteStop on a frozen policy with a single all-mask forward means termination can be retuned or rethresholded without retraining the motion policy, which is directly useful for deployment tuning.
The distance-wise ablations imply the three components address separable failure modes: LiteStop matters near the goal, memory matters at intermediate distances, and chunk planning provides lookahead on long trajectories; a practitioner can therefore prioritize components by operating range.

Reading between the lines

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

I would expect the read-before-write memory recipe to transfer to other partially observable navigation tasks, including ground-based VLN; a cheap test would be to swap the aerial backbone for a ground policy and measure whether the same ablation gap appears.
The LiteStop pattern suggests a general calibration trick for diffusion VLAs: the initial all-mask logits already encode the model's planning prior, so they can be repurposed as features for binary heads such as task completion or unsafe-state detection without extra forward passes.
A direct covariate-shift probe would be to run closed-loop navigation while feeding expert-prefix memories instead of online memories; if SR stays near the reported values, the memory mechanism is robust to distribution shift, and if it collapses, the gains partly reflect train/test memory mismatch.
Because evaluation is simulation-only, a natural next step is physical UAV flight; the main uncertainty is whether memory slots derived from frozen visual routers survive real sensing noise, which the current benchmark cannot answer.

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.

Desk Editor's Note

read the letter

Quick take: this is a sensible engineering paper with one real flaw in the headline comparison. The components are mostly known separately, but the combination — causal read-before-write memory, plan-K execute-one diffusion planning, and a separate LiteStop head — is new in aerial VLN, and the ablations are internally consistent. The paper is also honest about its main limitation: Section 3.3 admits training uses expert-prefix memories while deployment uses online memories that may differ, and Section 1 explicitly says covariate shift is not mitigated. That is a genuine concern, but it is measurable and fixable.

The bigger problem is the OpenFly-Agent baseline. Section 4.1.1 removes the pre-packaged historical keyframes and keeps only the current RGB observation. OpenFly-Agent is explicitly a keyframe-selection method, and Section 4.2 evaluates the official checkpoint under this stripped protocol. That is not a neutral standardization; it removes the input the baseline's design is built around. So the "outperforms all compared methods" claim is not supported as stated. There is also a smaller internal inconsistency: Table 2 reports DreamFly at 32.04% SR / 28.22% SPL, but Table 3 reports 31.46% / 27.17% for the same full model, with no explanation. That needs fixing before the numbers can be trusted.

To be fair, the paper does several things well. The memory formulation has a clean temporal boundary and the read-before-write protocol avoids future leakage. The valid-prefix supervision and horizon weighting in Section 3.4 are sensible, and the ablation patterns — memory helps at mid distances, LiteStop helps near the goal — are plausible. The LiteStop trick, reading the all-mask logit grid from a frozen policy, is neat and cheap. The qualitative examples are illustrative rather than evidence on their own, and the citations cover the relevant prior work.

What I would want before accepting the SOTA claim: a properly configured OpenFly-Agent baseline that receives its keyframe inputs, or an explicit argument for why stripping them is harmless (it probably is not); one consistent set of numbers for the full model; and ideally released code or a closed-loop study that measures the train/deployment memory mismatch. The covariate-shift issue is secondary to the baseline fairness issue.

Who is this for? People working on aerial VLN and memory-augmented VLA policies. It deserves a serious referee — not a desk reject — because the questions are real and the design is well motivated. I would send it out with the baseline-comparison concern flagged up front. My own verdict: conditional, leaning reject as-is because the headline result is not backed by the evidence, but it is a short path to a decent paper.

Referee Report

2 major / 4 minor

Summary. DreamFly is an aerial vision-language navigation system built on the Dream-VLA diffusion backbone. It contributes three components: a causally aligned historical memory that exposes only pre-decision observations to the policy via a read-before-write protocol; receding-horizon diffusion planning that predicts a K-step action chunk but executes only the first action; and LiteStop, a separate termination head trained on the frozen policy's initial all-mask action logits. Experiments on the OpenFly benchmark report the best SR, SPL, and NE among six baselines on test-seen and test-unseen splits, together with progressive and leave-one-out ablations and qualitative trajectory analyses.

Significance. If the comparison were properly configured, the paper would make a useful empirical contribution. The strengths are the explicit temporal boundary in memory construction, the plan-K/execute-one protocol, the decoupled termination objective, and the use of a disjoint calibration set for the LiteStop threshold. The implementation details are specific enough to allow replication of the architecture. However, the headline claim of outperforming all compared methods is currently not supported because the strongest learned baseline is evaluated under a protocol that removes its required keyframe history, and because the full-model numbers differ between Table 2 and Table 3. The underlying design remains plausible, but the quantitative evidence needs to be rerun or reconciled before the central claim can be accepted.

major comments (2)

[§4.1.1 and §4.2] The OpenFly-Agent baseline is evaluated from its official checkpoint under the standardized protocol described in §4.1.1, which removes the pre-packaged historical keyframes and retains only the current RGB observation at each decision step. OpenFly-Agent is described in §2.1 as a method that selects informative keyframes from long observation histories, so removing that input is not a neutral protocol but an ablation of the baseline's required modality. To support the claim of outperforming all compared methods, the authors must either evaluate the official OpenFly-Agent checkpoint with its intended keyframe input, or retrain OpenFly-Agent under their standardized protocol, and report both configurations if a discrepancy remains.
[Table 2 versus Table 3] The full DreamFly model is reported with different numbers in the two tables. Table 2 gives NE 44.87 m, SR 32.04%, and SPL 28.22% on test-seen, while Table 3's DreamFly(Ours) row gives NE 44.97 m, SR 31.46%, and SPL 27.17% without a split label. The abstract and conclusion cite the Table 2 values. The paper must state which split Table 3 reports, why the same configuration yields different numbers, and which checkpoint corresponds to the headline result; without this reconciliation the reader cannot determine which numbers support the central claim.

minor comments (4)

[§4.3] The ablation table reports a single run for each configuration with no variance or repeated-seed information; given that several ablation gaps are small (for example, the difference between the full model and the w/o Chunk variant in SPL), at least two seeds or an explicit statement that only one seed was used would help assess stability.
[§3.4 and §4.1.3] The CAR context coefficient is defined in §3.4, but §4.1.3 introduces a CAR reweighting probability p = 0.1 without explaining how it is applied during training; please clarify the stochastic application of CAR.
[Table 2] The abstract says DreamFly outperforms all compared methods on 'both metrics' and attains the lowest navigation error, but Table 2 shows that DreamFly's OSR is lower than several baselines on both splits; please state explicitly which metrics are included in the claim.
[Throughout] There are several proofreading issues, including inconsistent capitalization after display equations, 'diﬀiculties' in the Introduction, and 'oﬀicially' in §4.2; a careful proofreading pass is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: DreamFly is an empirical system trained with supervised objectives and evaluated on a held-out benchmark; the main concerns are comparison fairness and reporting consistency, not derivation loops.

full rationale

DreamFly's claims are empirical rather than derivational. The action policy is trained with the horizon-aware cross-entropy objective Lact over discrete action tokens, LiteStop is trained with the separate binary stop objective Lstop, and both are evaluated on held-out OpenFly splits. No component is defined in terms of the headline SR/SPL metrics, and no fitted parameter is renamed as a prediction: the chunk length K, horizon decay, and LiteStop threshold are selected on a pilot/calibration set described as 'disjoint from the final evaluation split.' The causal memory design is explicitly not claimed to solve covariate shift: Section 3.3 states that the temporal notion of causality 'does not mitigate the covariate shift arising from policy execution' and that 'the resulting memory states may differ' between expert-prefix training and online deployment, which is an acknowledged limitation rather than a hidden loop. The only substantive concerns in the paper are outside circularity: the OpenFly-Agent baseline is evaluated without the historical keyframes its design consumes, and the full-model numbers differ between Table 2 and Table 3. These are protocol fairness and reporting issues, not cases where an output equals an input by construction or where a load-bearing claim reduces to a self-citation. I therefore find no significant circularity.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The empirical claim rests on hand-chosen hyperparameters and several domain assumptions about data remapping, frozen router quality, and expert-to-online memory transfer. No new physical or mathematical entities are postulated; LiteStop is a learned module rather than an invented entity. No code or data artifacts are released that would independently substantiate these assumptions.

free parameters (6)

Action chunk horizon K = 4
Length of the predicted action chunk; chosen by the authors, no sensitivity analysis reported.
Horizon weighting gamma = 0.7
Exponential decay used to weight near-term action slots in the action loss Lact.
CAR context coefficient gamma_car = 0.1
Controls the geometric kernel that computes context coefficients in the action objective.
LiteStop threshold tau_stop = 0.50
Selected on a 64-trajectory calibration set to minimize the OSR-SR gap while preserving SR.
LiteStop positive class weight = 4.0
Fixed weight for stop-positive samples in the LiteStop loss Lstop.
Number of memory slots = 16
Fixed capacity of the long-term historical memory, chosen by design without reported variation.

assumptions (4)

domain assumption Remapping approximately 190,000 non-standard action labels, with -1 mapped to Go Up and -2 to Go Down, preserves the intended expert behavior.
All training and baselines use the modified labels, but the paper does not validate that the remapping matches the original semantic intent of the OpenFly data.
domain assumption Frozen CLIPSeg and OWLv2 routers with overlapping instruction token windows reliably extract instruction-relevant visual candidates from aerial egocentric views.
The memory construction depends entirely on these frozen routers; no failure analysis or quantitative check of candidate quality is provided.
domain assumption Memories built from expert prefix trajectories during training transfer to online memories built from the agent's own rollouts at deployment.
Section 3.3 explicitly states the resulting memory states may differ, and Section 1 disclaims mitigation of covariate shift. The closed-loop results depend on this transfer.
standard math The discrete diffusion sampling process inherited from Dream-VLA, with the monotonic origin sampler, produces executable action chunks without additional correctness guarantees.
The paper relies on the diffusion and denoising framework of Dream-VLA as a black-box sampler; no independent formal or empirical verification is included in this paper.

how reviews work

0 comments

Cite this review

Pith. "Pith review of DreamFly: Causal Memory and Receding-Horizon Diffusion Planning for Aerial Vision-Language Navigation." pith.science (2026). https://pith.science/paper/TA5IDNNQ

@misc{pith2026260812308,
  author       = {Pith},
  title        = {Pith review of: DreamFly: Causal Memory and Receding-Horizon Diffusion Planning for Aerial Vision-Language Navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TA5IDNNQ}},
  note         = {Machine review of arXiv:2608.12308}
}

abstract

Aerial vision-language navigation (VLN) requires an embodied agent to integrate visual evidence over time, plan future actions, and determine when it has reached a navigation goal under partial observability. Although recent VLA models offer a promising perception-to-action paradigm, adapting them to aerial navigation remains challenging due to limited historical context, short planning horizons, and unreliable implicit termination. To address these challenges, we propose DreamFly, a diffusion-based aerial VLN framework built on Dream-VLA. DreamFly introduces a causally aligned historical memory that augments the current visual representation using only observations preceding the current decision step, enabling temporal reasoning without future information leakage. We further formulate navigation as receding-horizon diffusion planning, where the policy predicts a $K$-step action chunk but executes only the first action before replanning. This plan-$K$, execute-one strategy uses future actions as auxiliary planning targets while preserving closed-loop visual feedback. Finally, LiteStop estimates the stop probability directly from action logits at the initial all-mask state, decoupling explicit termination from action generation. Experiments on the OpenFly benchmark demonstrate consistent improvements in seen and unseen environments. DreamFly achieves 32.04%/29.46% SR and 28.22%/23.54% SPL on the test-seen/test-unseen splits, respectively, outperforming all compared methods on both metrics while attaining the lowest navigation error. These results demonstrate the effectiveness of jointly modeling historical context, future action structure, and explicit termination for aerial VLN.

Figures

Figures reproduced from arXiv: 2608.12308 by the authors.

**Figure 1.** DreamFly at a Glance. At decision step t, DreamFly conditions on the language instruction, current aerial observation, and a historical memory containing only observations acquired before step t to generate a K-step action plan. DreamFly executes only the first action, acquires a new observation, and replans at the next decision step, while LiteStop separately estimates the probability of termination. Dashed action… view at source ↗

**Figure 2.** Overview of the proposed DreamFly framework. At each decision step, DreamFly conditions action prediction on the current observation, navigation instruction, and a causally aligned memory constructed from prior observations. The Dream-VLA backbone predicts a K-step action chunk through bidirectional diffusion, while LiteStop estimates termination from the initial all-mask action logits. If termination is not trigge… view at source ↗

**Figure 3.** Dataset statistics: (a) action distribution in the training set; (b) distributions of initial goal distances on the test-seen and test-unseen splits. 4.1.2 Evaluation Metrics. Following the evaluation protocol of OpenFly and prior aerial VLN studies, we adopt four standard metrics: navigation error (NE), success rate (SR), oracle success rate (OSR), and success weighted by path length (SPL). NE measures the average … view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Conditional success rates of baseline methods under different initial goal distances. Trajectories are partitioned according to whether the initial goal distance is within the 20 m success radius. The stochastic baselines exhibit substantially higher success rates whe…

**Figure 5.** Figure 5: Component-wise analysis across different initial 3D Euclidean distance(s) to the goal. Test trajectories are partitioned into three groups according to their initial shortest-path distances to examine how the contribution of each DreamFly component varies with navigati…

**Figure 6.** Figure 6: Qualitative comparison between DreamFly and its ablated variants on representative OpenFly trajectories. The three examples illustrate the effects of historical memory, recedinghorizon action planning, and LiteStop, respectively. Numbers indicate the distance to the n…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 14 canonical work pages

[1]

Vision-and-Language Navigation: Interpreting Visually-Grounded Nav- igation Instructions in Real Environments,

P. Anderson et al., “Vision-and-Language Navigation: Interpreting Visually-Grounded Nav- igation Instructions in Real Environments,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition , Jun. 2018, pp. 3674–3683. doi: 10.1109/CVPR.2018.00387. 20 Figure 6: Qualitative comparison between DreamFly and its ablated variants on representative...

arXiv 2018
[3]

Aerial Vision-and-Language Navigation with Grid-based View Selection and Map Construction

G. Zhao, G. Li, J. Pan, and Y. Yu. “Aerial Vision-and-Language Navigation with Grid-based View Selection and Map Construction. ” arXiv: 2503.11091 [cs.CV], pre-published

arXiv
[4]

OpenFly: A Comprehensive Platform for Aerial Vision-Language Navigation,

Y. Gao et al., “OpenFly: A Comprehensive Platform for Aerial Vision-Language Navigation,” in The Fourteenth International Conference on Learning Representations , 2026

work page 2026
[5]

Reinforced Cross-Modal Matching and Self-Supervised Imitation Learning for Vision-Language Navigation,

X. Wang et al., “Reinforced Cross-Modal Matching and Self-Supervised Imitation Learning for Vision-Language Navigation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2019, pp. 6629–6638

work page 2019
[7]

History Aware Multimodal Transformer for Vision-and-Language Navigation,

S. Chen, P.-L. Guhur, C. Schmid, and I. Laptev, “History Aware Multimodal Transformer for Vision-and-Language Navigation,” in Advances in Neural Information Processing Systems , M. Ranzato, A. Beygelzimer, Y. Dauphin, P. S. Liang, and J. W. Vaughan, Eds., vol. 34, Curran Associates, Inc., 2021, pp. 5834–5847

work page 2021
[8]

LongFly: Long-Horizon UA V Vision-and-Language Navigation with Spa- tiotemporal Context Integration,

W. Jiang et al., “LongFly: Long-Horizon UA V Vision-and-Language Navigation with Spa- tiotemporal Context Integration,” arXiv:2512.22010, 2025. [Online]. A vailable: https://arxiv. org/abs/2512.22010

arXiv 2025
[9]

RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control,

B. Zitkovich et al., “RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control,” in Proceedings of The 7th Conference on Robot Learning , PMLR, Dec. 2, 2023, pp. 2165–2183

work page 2023
[10]

OpenVLA: An Open-Source Vision-Language-Action Model,

M. J. Kim et al., “OpenVLA: An Open-Source Vision-Language-Action Model,” in Proceedings of the 8th Conference on Robot Learning , vol. 270, PMLR, 2025, pp. 2679–2713

work page 2025

Show all 31 references

[11]

Diffusion policy: Visuomotor policy learning via action diffusion,

C. Chi et al., “Diffusion policy: Visuomotor policy learning via action diffusion,” The International Journal of Robotics Research , vol. 44, no. 10–11, pp. 1684–1704, 2025. doi: 10.1177/02783649241273668

2025 doi
[12]

DreamVLA: A Vision-Language-Action Model Dreamed with Comprehen- sive World Knowledge,

W. Zhang et al., “DreamVLA: A Vision-Language-Action Model Dreamed with Comprehen- sive World Knowledge,” in Advances in Neural Information Processing Systems 38 , 2025

2025
[13]

AerialVLA: A Vision-Language-Action Model for UA V Navigation via Minimalist End-to-End Control,

P. Xu, Z. Deng, J. Deng, Z. Gu, and S. Wan, “AerialVLA: A Vision-Language-Action Model for UA V Navigation via Minimalist End-to-End Control,” arXiv:2603.14363, 2026. [Online]. A vailable:https://arxiv.org/abs/2603.14363

2026
[14]

WorldVLN: Autoregressive World Action Model for Aerial Vision-Language Navigation,

B. Zhao et al., “WorldVLN: Autoregressive World Action Model for Aerial Vision-Language Navigation,” arXiv:2605.15964, 2026. [Online]. A vailable: https://arxiv.org/abs/2605.15964

2026 arXiv
[15]

ImagineUA V: Aerial Vision-Language Navigation via World-Action Modeling and Kinodynamic Planning,

X. Liu, J. Huang, S. Xia, B. Liu, J. Cui, and J. Yang, “ImagineUA V: Aerial Vision-Language Navigation via World-Action Modeling and Kinodynamic Planning,” arXiv:2606.01205, 2026. [Online]. A vailable:https://arxiv.org/abs/2606.01205

2026 arXiv
[16]

FSD-VLN: Fast-Slow Dual-System Modeling for Aerial Long-Horizon Vision- Language Navigation,

X. Zhu et al., “FSD-VLN: Fast-Slow Dual-System Modeling for Aerial Long-Horizon Vision- Language Navigation,” arXiv:2607.08359, 2026. [Online]. A vailable: https://arxiv.org/abs/ 2607.08359 22

2026 arXiv
[17]

A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning,

S. Ross, G. Gordon, and D. Bagnell, “A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning,” in Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics , vol. 15, PMLR, 2011, pp. 627–635

2011
[18]

Learning to Stop: A Simple yet Effective Approach to Urban Vision-Language Navigation,

J. Xiang, X. Wang, and W. Y. Wang, “Learning to Stop: A Simple yet Effective Approach to Urban Vision-Language Navigation,” in Findings of the Association for Computational Linguistics: EMNLP 2020 , Association for Computational Linguistics, 2020, pp. 699–707. doi: 10.18653/v1...

2020 doi
[19]

Aerial Vision-and-Dialog Navigation

Y. Fan, W. Chen, T. Jiang, C. Zhou, Y. Zhang, and X. E. Wang. “Aerial Vision-and-Dialog Navigation. ” arXiv:2205.12219 [cs.CV], pre-published

arXiv
[20]

CityNav: A Large-Scale Dataset for Real-World Aerial Navigation

J. Lee et al. “CityNav: A Large-Scale Dataset for Real-World Aerial Navigation. ” arXiv: 2406. 14240 [cs.CV], pre-published
[21]

Towards Realistic UA V Vision-Language Navigation: Platform, Benchmark, and Methodology

X. Wang et al. “Towards Realistic UA V Vision-Language Navigation: Platform, Benchmark, and Methodology. ” arXiv: 2410.07087 [cs.CV], pre-published

arXiv
[22]

AirNav: A Large-Scale UA V Vision-and-Language Navigation Dataset with Natural and Diverse Instructions

H. Cai et al. “AirNav: A Large-Scale UA V Vision-and-Language Navigation Dataset with Natural and Diverse Instructions. ” arXiv: 2601.03707 [cs.CL], pre-published

arXiv
[23]

Exploring Spatial Representation to Enhance LLM Reasoning in Aerial Vision-Language Navigation

Y. Gao, Z. Wang, P. Han, L. Jing, D. Wang, and B. Zhao. “Exploring Spatial Representation to Enhance LLM Reasoning in Aerial Vision-Language Navigation. ” arXiv: 2410.08500 [cs.RO], pre-published

arXiv
[24]

History-Enhanced Two-Stage Transformer for Aerial Vision-and-Language Navigation

X. Ding, J. Gao, C. Pan, W. Wang, and J. Qin. “History-Enhanced Two-Stage Transformer for Aerial Vision-and-Language Navigation. ” arXiv: 2512.14222 [cs.CV], pre-published
[25]

LookasideVLN: Direction-Aware Aerial Vision-and-Language Navigation,

Y. Ning et al., “LookasideVLN: Direction-Aware Aerial Vision-and-Language Navigation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2026, pp. 32 441–32 450

2026
[26]

FlightGPT: Towards Generalizable and Interpretable UA V Vision-and- Language Navigation with Vision-Language Models,

H. Cai et al., “FlightGPT: Towards Generalizable and Interpretable UA V Vision-and- Language Navigation with Vision-Language Models,” in Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , C. Christodoulopoulos, T. Chakraborty, C. Rose, and...

2025 doi
[27]

FineCog-Nav: Integrating Fine-grained Cognitive Modules for Zero-shot Mul- timodal UA V Navigation,

D. Shao et al., “FineCog-Nav: Integrating Fine-grained Cognitive Modules for Zero-shot Mul- timodal UA V Navigation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings , 2026, pp. 1325–1334

2026
[28]

Structured Scene Memory for Vision- Language Navigation,

H. Wang, W. Wang, W. Liang, C. Xiong, and J. Shen, “Structured Scene Memory for Vision- Language Navigation,” in 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2021, pp. 8451–8460. doi: 10.1109/CVPR46437.2021.00835

2021
[29]

GridMM: Grid Memory Map for Vision-and- Language Navigation,

Z. Wang, X. Li, J. Yang, Y. Liu, and S. Jiang, “GridMM: Grid Memory Map for Vision-and- Language Navigation,” in 2023 IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France: IEEE, Oct. 1, 2023, pp. 15 579–15 590, isbn: 979-8-3503-0718-4. doi: 10.1109/ICCV51...

2023
[30]

Learning Fine-Grained Bimanual Manipula- tion with Low-Cost Hardware,

T. Z. Zhao, V. Kumar, S. Levine, and C. Finn, “Learning Fine-Grained Bimanual Manipula- tion with Low-Cost Hardware,” in Proceedings of Robotics: Science and Systems XIX , Daegu, Republic of Korea, 2023. doi: 10.15607/RSS.2023.XIX.016

2023 doi
[31]

Bidirectional Decoding: Improv- ing Action Chunking via Guided Test-Time Sampling,

Y. Liu, J. I. Hamid, A. Xie, Y. Lee, M. Du, and C. Finn. “Bidirectional Decoding: Improv- ing Action Chunking via Guided Test-Time Sampling,” arXiv.org, Accessed: Aug. 12, 2026. [Online]. A vailable:https://arxiv.org/abs/2408.17355v4 23

2026 arXiv
[32]

Real-Time Execution of Action Chunking Flow Policies,

K. Black, M. Galliker, and S. Levine, “Real-Time Execution of Action Chunking Flow Policies,” in Advances in Neural Information Processing Systems , vol. 38, Curran Associates, Inc., 2025, pp. 33 383–33 407. doi: 10.52202/085713-1122

2025 doi
[33]

Dream-VL & Dream-VLA: Open Vision-Language and Vision-Language-Action Models with Diffusion Language Model Backbone

J. Ye et al. “Dream-VL & Dream-VLA: Open Vision-Language and Vision-Language-Action Models with Diffusion Language Model Backbone. ” arXiv: 2512.22615 [cs.CV], pre-published. 24

Pith tools

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

[1] [1]

Vision-and-Language Navigation: Interpreting Visually-Grounded Nav- igation Instructions in Real Environments,

P. Anderson et al., “Vision-and-Language Navigation: Interpreting Visually-Grounded Nav- igation Instructions in Real Environments,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition , Jun. 2018, pp. 3674–3683. doi: 10.1109/CVPR.2018.00387. 20 Figure 6: Qualitative comparison between DreamFly and its ablated variants on representative...

arXiv 2018

[2] [3]

Aerial Vision-and-Language Navigation with Grid-based View Selection and Map Construction

G. Zhao, G. Li, J. Pan, and Y. Yu. “Aerial Vision-and-Language Navigation with Grid-based View Selection and Map Construction. ” arXiv: 2503.11091 [cs.CV], pre-published

arXiv

[3] [4]

OpenFly: A Comprehensive Platform for Aerial Vision-Language Navigation,

Y. Gao et al., “OpenFly: A Comprehensive Platform for Aerial Vision-Language Navigation,” in The Fourteenth International Conference on Learning Representations , 2026

work page 2026

[4] [5]

Reinforced Cross-Modal Matching and Self-Supervised Imitation Learning for Vision-Language Navigation,

X. Wang et al., “Reinforced Cross-Modal Matching and Self-Supervised Imitation Learning for Vision-Language Navigation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2019, pp. 6629–6638

work page 2019

[5] [7]

History Aware Multimodal Transformer for Vision-and-Language Navigation,

S. Chen, P.-L. Guhur, C. Schmid, and I. Laptev, “History Aware Multimodal Transformer for Vision-and-Language Navigation,” in Advances in Neural Information Processing Systems , M. Ranzato, A. Beygelzimer, Y. Dauphin, P. S. Liang, and J. W. Vaughan, Eds., vol. 34, Curran Associates, Inc., 2021, pp. 5834–5847

work page 2021

[6] [8]

LongFly: Long-Horizon UA V Vision-and-Language Navigation with Spa- tiotemporal Context Integration,

W. Jiang et al., “LongFly: Long-Horizon UA V Vision-and-Language Navigation with Spa- tiotemporal Context Integration,” arXiv:2512.22010, 2025. [Online]. A vailable: https://arxiv. org/abs/2512.22010

arXiv 2025

[7] [9]

RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control,

B. Zitkovich et al., “RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control,” in Proceedings of The 7th Conference on Robot Learning , PMLR, Dec. 2, 2023, pp. 2165–2183

work page 2023

[8] [10]

OpenVLA: An Open-Source Vision-Language-Action Model,

M. J. Kim et al., “OpenVLA: An Open-Source Vision-Language-Action Model,” in Proceedings of the 8th Conference on Robot Learning , vol. 270, PMLR, 2025, pp. 2679–2713

work page 2025