OnEvoMemory: Evolving Memory through Online Robot Rollouts for Pretrained Robot Policies

Shenqi Zong; Zhongxi Chen

REVIEW 5 major objections 5 minor 15 references

OnEvoMemory: Evolving Memory through Online Robot Rollouts for Pretrained Robot Policies

T0 review · 5 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A value-guided memory module, seeded from demonstrations and refined by successful and failed robot rollouts, claims to make a frozen pretrained policy better at long-horizon manipulation.

desk verdict Plausible memory module for frozen VLA policies, but the unspecified value training and thin, seed-free evaluation leave the central claim unproven. read the letter →

arxiv 2608.08749 v1 pith:4ZIUUVCC submitted 2026-08-09 cs.RO

Zhongxi Chen , Shenqi Zong This is my paper · ORCID

classification cs.RO

keywords long-horizonmanipulationvision-language-actionmodelsvalue-guidedmemoryonlineevolutionhierarchicalrobotrolloutsfrozenpolicyadaptationofflinedemonstrationinitialization

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

OnEvoMemory claims that a pretrained vision-language-action robot policy can become better at long-horizon manipulation without any modification to the policy itself, by adding a memory module that learns what to remember. The module keeps a short-term raw buffer, an elite bank of high-value experiences, and a transition bank of moments where trajectory value changes sharply; a learned value estimator decides what goes into each bank. Offline demonstrations initialize the writer, and successful plus failed online rollouts refine it. On the paper's benchmarks, offline memory initialization raises average success on LiberoLong-10 from 86.2% to 88.6%, and one round of online memory evolution raises it further to 90.2%, while two RMBench tasks move from 0% success to 14% and 10%. If true, the point matters because memory selection is usually hand-designed or outsourced to external models, whereas this work tries to learn it from interaction outcomes while keeping the policy frozen.

What carries the argument

The central mechanism is the action-conditioned value estimator $(V_t, k_t, v_t) = F_\phi(\tilde{Q}_t, A_t)$, operating on action-query representations $Q_t = E_\theta(o_t)$ from the frozen VLA. Its writing score $V_t$ places high-value experiences into an elite bank and large temporal value changes $|V_t - V_{t-1}|$ into a transition bank, while a fixed-capacity FIFO short-term buffer preserves recent context; retrieval returns top-k similar plus recently written items from each bank, and gated cross-attention fuses that context into the frozen action decoder. This single learned writer replaces fixed temporal windows, hand-written event rules, and external models for deciding what history is worth keeping.

What would settle it

Run the same experiments with the value-guided writer ablated: fill the elite and transition banks with random or uniformly sampled experiences while keeping retrieval and gated cross-attention identical. If the reported success-rate gains on LiberoLong-10 and RMBench largely persist, then the value estimator's scoring is not the cause and the central claim about learned memory selection collapses.

Watch

Extended reading notes

Core claim

The central discovery is that memory selection for long-horizon manipulation can be learned from trajectory outcomes and injected into a frozen policy. In the paper's architecture, action-query representations $Q_t = E_\theta(o_t)$ from the frozen VLA are read against three memory banks through gated cross-attention; a value estimator $F_\phi$ scores each experience and routes high-value items to an elite bank and large temporal value changes to a transition bank, while retrieval combines top-k similarity with recently written items. The paper reports that this structure improves the base policy on LiberoLong-10 from 86.2% to 88.6% with offline-initialized memory and to 90.2% after online evolution, and that it gives the first successes on RMBench's SwapBlocks and SwapT, moving both from 0% to 14% and 10%. The authors interpret these results as evidence that a value-guided writer can recognize task-stage transitions and keep the policy from repeating completed subtasks without changing the pretrained action decoder.

Load-bearing premise

The method works only if the frozen policy's action-query vectors carry enough information about task state and outcome quality that the learned value estimator can tell which past moments are worth keeping and when the task stage changed.

Editorial extensions

If this is right

Attaching OnEvoMemory to a frozen VLA can improve long-horizon success without retraining the policy, making memory adaptation a separable axis from policy adaptation.
Both successful and unsuccessful rollouts carry usable supervision for memory selection; failure trajectories are not discarded but are used to revise overvalued experiences.
Large temporal changes in the learned value score can mark task-stage boundaries, grasp success, and contact loss, giving the policy a way to track progress without external event detection.
One round of online memory evolution with ten to twenty trajectories per task already moves success rates upward, suggesting memory quality can be improved cheaply relative to policy fine-tuning.

Reading between the lines

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

The design implies a testable decomposition: replace the value-guided writer with random or heuristic writing while keeping the banks and retrieval fixed; if the gains persist, the value estimator is not the active ingredient.
Because only the memory modules are updated, OnEvoMemory could in principle be attached to other token-based VLA backbones; the paper's numbers alone do not show whether the learned value signal transfers across backbones.
The one-round online protocol leaves open whether additional rollout rounds continue to improve memory or saturate; a multi-round experiment would map the evolution curve.
Since the transition bank records jumps in value, the same writer could be re-purposed to auto-label demonstration keyframes, turning OnEvoMemory into a data-curation tool rather than only a policy add-on.

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.

Desk Editor's Note

read the letter

Here's the thing you need to know: OnEvoMemory is a plausible memory module for frozen VLA policies, but the paper leaves the value-guided part underspecified and the evaluation is too thin to carry the claim.

What's genuinely new is the combination of a hierarchical memory (short-term buffer, elite bank, transition bank) with a learned value estimator that decides what to write, plus offline initialization from demonstrations and online refinement from rollouts. Keeping the base policy frozen is a real practical advantage. The writing is clear, the motivation (tracking completed subtasks, grasping and contact events) makes sense, and the authors are honest that the gains are moderate and RMBench rates are still low.

The soft spots are substantial, though. The biggest is that Section 3.1 defines the value estimator F_phi, but Section 3.2 never gives its loss or training signal. Saying 'action supervision and trajectory outcomes jointly initialize' isn't enough. Without knowing how episode success becomes per-timestep value targets, the reported gains could come from any retrieval mechanism. The natural ablation—swap the value-guided writer with a fixed or heuristic writer—is missing. No baseline comparison with MemoryWAM, KEMO, HAMLET, EventVLA, or similar memory policies, even though they're cited. The stress-test note is right on this.

The numbers are also reported loosely. Table 1 has no variance, no seed count, and no statistical tests. On RMBench you collect 10 trajectories per task, but report 14% success; that's not an integer unless you averaged over multiple seeds, which isn't stated. Small absolute gains (0 to 14% on SwapBlocks) could be one or two lucky rollouts. I'd want to see error bars and per-seed counts.

I don't think the central idea is wrong; it's just not yet supported. The paper reads like a preliminary preprint that needs a full version with ablations, baselines, and proper experimental detail. Who is this for? People working on memory-augmented VLA policies will want to know about it, and the frozen-policy setup is relevant to anyone deploying pretrained policies in long-horizon tasks. I'd send it to peer review, but with the expectation of major revision. A serious reviewer could quickly turn this into a solid paper by demanding the missing loss function and a couple of ablations.

Referee Report

5 major / 5 minor

Summary. The paper proposes OnEvoMemory, a value-guided memory module that augments a frozen pretrained vision-language-action (VLA) policy for long-horizon robot manipulation. The memory consists of three banks: a short-term FIFO buffer, an elite experience bank, and a transition bank. Reading is performed by retrieving a union of top-k similar and recently written entries from the long-term banks, which are injected via gated cross-attention into the policy's action-query representations. Writing is guided by a learned value estimator F_phi that outputs a writing score V_t, a retrieval key k_t, and a stored representation v_t; high-value experiences go to the elite bank and large temporal value differences go to the transition bank. The module is first initialized by replaying offline demonstrations, then refined through one round of online rollouts using both successful and unsuccessful trajectories, while the base policy remains frozen. Experiments on LiberoLong-10 and two RMBench tasks report higher success rates with offline memory initialization and further gains after online evolution.

Significance. If the reported results are robust, the paper's central contribution is a plausible new mechanism for adapting memory selection to a frozen VLA policy without modifying the policy itself. The architecture is clearly described at a high level, and the decision to keep the policy frozen during online adaptation is a practically valuable design choice that addresses a real deployment constraint. The paper also deserves credit for explicitly acknowledging the moderation of its gains and the limited absolute success rates on RMBench, which is more transparent than is typical. However, the significance is currently limited because the experimental evidence is thin, the value-guided writing mechanism is not fully specified, and no comparison is made against the memory baselines cited in the related work. The claim that value guidance rather than generic retrieval is the active ingredient is not yet established.

major comments (5)

[Section 3.2] The training procedure for the value estimator and memory modules is not specified. The text says that 'action supervision and trajectory outcomes jointly initialize the memory prior and train the memory reading, writing, and injection modules,' but no loss function, no per-timestep value target, and no optimization objective are given. Since the entire writing mechanism depends on V_t aligning with grasp success, contact loss, and task-stage transitions, the absence of a concrete training objective makes the method non-reproducible and leaves open the possibility that V_t encodes recency or similarity rather than task-relevant value.
[Table 1] The experimental evidence consists of a single table with no variance, no number of seeds, no number of evaluation episodes, and no statistical test. The LiberoLong-10 gains are 2.4 points offline and 1.6 points online, which could easily be within rollout noise. The RMBench numbers are based on only 10 online adaptation trajectories and 0% baseline success; without confidence intervals or multiple seeds, the claimed improvements are not convincingly supported.
[Section 4] The online evaluation protocol is underspecified and potentially circular. The paper states that 20 rollout trajectories per task on LiberoLong-10 and 10 per task on RMBench are collected for online adaptation, but it does not state whether the reported '+ Online' success rates are measured on fresh evaluation episodes or on the same rollouts used for adaptation. If evaluation is performed on the adaptation rollouts, the online improvement would be partly built into the reported numbers. This must be clarified and, if necessary, corrected with fresh-episode evaluation.
[Section 2 and Section 4] The paper criticizes existing memory mechanisms such as BPP, MemER, MemoryWAM, EventVLA, KEMO, and WeaveLA, but it never compares against any of them. Without at least one related baseline evaluated under the same base policy and evaluation protocol, the paper's claim that OnEvoMemory improves over 'existing memory mechanisms' is unsupported. A minimum requirement is to run a representative baseline, such as a fixed similarity-based retrieval or a manually defined keyframe selector, under identical conditions.
[Section 3.1] There is no ablation that isolates the value-guided writing mechanism. A natural control would replace the learned value score V_t with a random score, a recency score, or a fixed heuristic (e.g., writing every k-th timestep), while keeping the memory banks and retrieval unchanged. Without such an ablation, the reported gains could be produced by the retrieval-augmented context alone, which would reduce the central contribution to a retrieval system rather than value-guided memory evolution.

minor comments (5)

[Section 3.1] The notation (V_t, k_t, v_t) = F_phi(Qtilde_t, A_t) is introduced but the roles of k_t and v_t are not fully elaborated; in particular, it is unclear whether k_t is derived from Qtilde_t, from A_t, or from both, and whether v_t is a learned projection of either. The paper should state the dimensionalities and the exact inputs to the retrieval key and stored representation.
[Section 3.2] The phrase 'trajectory outcomes' is used repeatedly but never defined operationally. It should be stated explicitly whether an outcome is a binary episode success, a sparse reward, a dense reward, or a combination, and how this outcome is assigned to individual timesteps for updating F_phi.
[Section 4] The paper does not report how many evaluation episodes were run per task for each setting, nor the variance across tasks on LiberoLong-10 beyond the average. Reporting per-task success rates with confidence intervals would substantially strengthen the empirical claims.
[Section 4] The sentence 'The additional gains from online adaptation may be attributed to the unsuccessful rollouts collected during interaction' is speculative; the paper provides no analysis separating the contribution of successful versus unsuccessful rollouts. A small experiment using only successful rollouts, or only unsuccessful rollouts, would make this attribution concrete.
[General] There are minor formatting issues, including the inconsistent use of 'OnEvoMemory' with and without a space in the abstract and introduction, and the reference to 'QwenOFT' cites two papers [3,11] that are about fine-tuning VLAs and a codebase, respectively, rather than a model named QwenOFT; the authors should cite the actual base model or clarify its provenance.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: OnEvoMemory is an empirical learning pipeline whose reported gains are not by-construction consequences of its inputs.

full rationale

OnEvoMemory does not offer a formal derivation, and its central claims are empirical: offline memory initialization and one round of online rollout adaptation improve success rates on LiberoLong-10 and RMBench over a frozen base VLA. Nothing in the method defines the output in terms of the input or fits a parameter and then renames it as a prediction. The value estimator F_phi is trained from trajectory outcomes; elite and transition banks are populated by learned writing scores and temporal value differences, which is a learning procedure, not a tautology. There are no self-citations from the authors that carry a load-bearing argument. The main weaknesses are incompleteness rather than circularity: Section 3.2 states that 'Action supervision and trajectory outcomes jointly initialize the memory prior and train the memory reading, writing, and injection modules' but gives no loss function or training objective, and the paper does not state whether the +Online success rates are measured on held-out episodes or on the same 20/10 trajectories used for adaptation. If the latter were true, the online gain would be partly circular, but the paper does not say this, the default evaluation protocol in this literature is held-out evaluation, and hard rule 1 prohibits speculating about an unstated protocol. The qualitative claim that large temporal changes in V_t correspond to task-stage transitions is also an empirical assertion about what the learned value encodes, not a definitional equivalence, because V_t is learned rather than hand-defined to equal task progress. Accordingly, no circular step can be established from the text, and the appropriate finding is no significant circularity.

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

The paper's empirical method depends on several unstated modeling choices: the informativeness of the frozen policy's action queries, the sufficiency of scalar success and failure outcome labels for learning value-based event salience, the adequacy of fixed TopK and RecentK retrieval, and the representativeness of a small online rollout budget. The free parameters are writing thresholds and memory capacities that are not reported.

free parameters (4)

elite-bank writing threshold
The value-guided writer sends high-value experiences to M^E when V_t exceeds an unstated threshold; no value, annealing schedule, or source is given in Section 3.1.
transition-bank writing threshold
Experiences are written to M^T when |V_t - V_{t-1}| is large; the definition of large is not specified in Section 3.1.
memory capacities and retrieval sizes
FIFO capacity, elite-bank capacity, transition-bank capacity, TopK and RecentK are not reported; all are hand-selected module hyperparameters in Section 3.1.
online adaptation hyperparameters (update steps, learning rate, stopping criteria) = 10 to 20 rollouts per task, one round
Section 4 gives only the rollout budget; training iterations, learning rates, and stopping criteria are omitted, so the online gains cannot be reproduced or attributed.

assumptions (4)

domain assumption The frozen base VLA exposes action-query representations Q_t that are informative as memory keys and values and as inputs to the value estimator.
Section 3 defines Q_t = E_theta(o_t) and uses it for both retrieval and value prediction, but no evidence is given that these queries encode stage-completion information.
domain assumption Trajectory outcome labels (success and failure) are sufficient supervision to train a value estimator whose temporal differences mark events such as grasp success, contact loss, and task-stage transitions.
Section 3.2 says outcomes initialize and refine the value and writing modules; no loss function or evidence connects scalar outcome labels to the required event salience.
domain assumption TopK similarity plus RecentK retrieval from the two long-term banks provides enough context for the gated cross-attention to improve long-horizon decisions.
Section 3.1 specifies retrieval as R_t = TopK(Q_t, M_b_t) union RecentK(M_b_t); the sufficiency of this fixed scheme is not tested by ablation.
domain assumption One round of 10 to 20 online rollouts per task is a representative adaptation signal.
Section 4 fixes the online budget; no sensitivity analysis shows that the result is stable under different online budgets.

how reviews work

0 comments

Cite this review

Pith. "Pith review of OnEvoMemory: Evolving Memory through Online Robot Rollouts for Pretrained Robot Policies." pith.science (2026). https://pith.science/paper/4ZIUUVCC

@misc{pith2026260808749,
  author       = {Pith},
  title        = {Pith review of: OnEvoMemory: Evolving Memory through Online Robot Rollouts for Pretrained Robot Policies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4ZIUUVCC}},
  note         = {Machine review of arXiv:2608.08749}
}

read the original abstract

Long-horizon robot manipulation requires policies to track completed subtasks and critical interaction events. However, existing memory mechanisms heavily rely on external models or predefined update rules. To address this, we propose OnEvoMemory, a value-guided memory module for pretrained robot policies. It maintains recent context, high-value experiences, and salient transitions, while learning which experiences should be retained from trajectory outcomes. Offline demonstrations initialize the memory prior, whereas successful and unsuccessful online rollouts refine memory selection, helping the policy recognize task-stage transitions and avoid repeating completed subtasks. Experiments on long-horizon manipulation benchmarks show that OnEvoMemory improves the performance of the base VLA policy through both offline initialization and online memory evolution.

Figures

Figures reproduced from arXiv: 2608.08749 by the authors.

**Figure 1.** Overview of OnEvoMemory. A frozen base VLA produces action-query representations, which are organized into short-term, elite, and transition memories through value-guided writing. Successful and unsuccessful robot rollouts further update only the value and memory modules, enabling online memory evolution without modifying the pretrained policy. – We introduce a hierarchical memory architecture consisting of a short… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 7 canonical work pages

[1]

Chen, T., Wang, Y., Li, M., Qin, Y., Shi, H., Li, Z., Hu, Y., Zhang, Y., Wang, K., Chen, Y., Wang, H., Xu, R., Wu, R., Mu, Y., Yang, Y., Dong, H., Luo, P.: Rmbench: Memory-dependent robotic manipulation benchmark with insights into policy design (2026),https://arxiv.org/abs/2603.01229

arXiv 2026
[2]

Chi, C., Xu, Z., Feng, S., Cousineau, E., Du, Y., Burchfiel, B., Tedrake, R., Song, S.: Diffusion policy: Visuomotor policy learning via action diffusion. Int. J. Robotics Res.44(10-11), 1684–1704 (2025).https : / / doi . org / 10 . 1177 / 02783649241273668,https://doi.org/10.1177/02783649241273668

doi:10.1177/02783649241273668 2025
[3]

In: Proceedings of Robotics: Science and Systems

Kim, M.J., Finn, C., Liang, P.: Fine-Tuning Vision-Language-Action Models: Op- timizing Speed and Success. In: Proceedings of Robotics: Science and Systems. LosAngeles, CA, USA (June 2025).https://doi.org/10.15607/RSS.2025.XXI. 017

doi:10.15607/rss.2025.xxi 2025
[4]

In: Agrawal, P., Kroemer, O., Burgard, W

Kim, M.J., Pertsch, K., Karamcheti, S., Xiao, T., Balakrishna, A., Nair, S., Rafailov, R., Foster, E.P., Sanketi, P.R., Vuong, Q., Kollar, T., Burchfiel, B., Tedrake, R., Sadigh, D., Levine, S., Liang, P., Finn, C.: Openvla: An open- source vision-language-action model. In: Agrawal, P., Kroemer, O., Burgard, W. (eds.) Conference on Robot Learning, 6-9 Nov...

work page 2025
[5]

In: The Fourteenth International Conference on Learning Representations (2026),https: //openreview.net/forum?id=KcJ9U0x6kO

Koo, M., Choi, D., Kim, T., Lee, K., Kim, C., Seo, Y., Shin, J.: HAMLET: Switch your vision-language-action model into a history-aware policy. In: The Fourteenth International Conference on Learning Representations (2026),https: //openreview.net/forum?id=KcJ9U0x6kO

work page 2026
[6]

In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S

Liu, B., Zhu, Y., Gao, C., Feng, Y., Liu, Q., Zhu, Y., Stone, P.: Libero: Bench- marking knowledge transfer for lifelong robot learning. In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S. (eds.) Advances in Neu- ral Information Processing Systems. vol. 36, pp. 44776–44791. Curran Associates, Inc. (2023),https://proceedings.neurips.c...

work page 2023
[7]

arXiv preprint arXiv:2602.15010 (2026),https://arxiv.org/abs/2602.15010

Mark, M.S., Liang, J., Attarian, M., Fu, C., Dwibedi, D., Shah, D., Kumar, A.: BPP: Long-context robot imitation learning by focusing on key history frames. arXiv preprint arXiv:2602.15010 (2026),https://arxiv.org/abs/2602.15010

arXiv 2026
[8]

In: Proceedings of Robotics: Science and Systems (2026)

Shah, R., Li, Y., Bello, F., Zhu, Y., Martín-Martín, R.: Memory retrieval in visuo- motor policies for long-horizon robot control. In: Proceedings of Robotics: Science and Systems (2026)

work page 2026

Show all 15 references

[9]

In: The Fourteenth International Con- ference on Learning Representations (2026),https://openreview.net/forum?id= 54U3XHf7qq

Shi, H., Xie, B., Liu, Y., Sun, L., Liu, F., Wang, T., Zhou, E., Fan, H., Zhang, X., Huang, G.: MemoryVLA: Perceptual-cognitive memory in vision-language- action models for robotic manipulation. In: The Fourteenth International Con- ference on Learning Representations (2026),h...

2026
[10]

In: The Fourteenth International Conference on Learning Representations (2026),https://openreview.net/forum?id=1dH4ARGdwD

Sridhar, A., Pan, J., Sharma, S., Finn, C.: Scaling up memory for robotic control via experience retrieval. In: The Fourteenth International Conference on Learning Representations (2026),https://openreview.net/forum?id=1dH4ARGdwD

2026
[11]

ArXivabs/2604.05014(2026),https://arxiv.org/abs/ 2604.05014

StarVLA Community: Starvla: A lego-like codebase for vision-language-action model developing. ArXivabs/2604.05014(2026),https://arxiv.org/abs/ 2604.05014

2026 arXiv
[12]

Chen and S

Yang, G., Tu, Z., Yang, Y., Mao, S., Dong, J., Chen, T., Peng, J., Xiong, J., Cao, J., Dai, J., Zhou, W., Mu, Y., Wang, T.: EventVLA: Event-driven visual 6 Z. Chen and S. Zong evidence memory for long-horizon vision-language-action policies. arXiv preprint arXiv:2606.20092 (20...

2026 arXiv
[13]

Yang, S., Mu, J., Wei, T., Lu, C., Li, X., Xu, L., Xue, Z., Yuan, Z., Lin, D., Pang, J., Xu, H.: Memorywam: Efficient world action modeling with persistent memory (2026),https://arxiv.org/abs/2606.20562

2026 arXiv
[14]

arXiv preprint arXiv:2606.23589 (2026),https://arxiv.org/abs/2606.23589

Zeng, Y., Ye, M., Chen, Y., Shentu, Y., Wu, P., Yan, Z., Li, Z.: KEMO: Event- driven keyframe memory for long-horizon robot manipulation with VLA policies. arXiv preprint arXiv:2606.23589 (2026),https://arxiv.org/abs/2606.23589

2026 arXiv
[15]

arXiv preprint arXiv:2606.17463 (2026),https://arxiv.org/abs/ 2606.17463

Zhu, S., Liu, Z., Wang, F., Wang, J., Yue, B., Liu, G., Wu, S., Xue, X., Zeng, T.: WeaveLA: Event driven cross-subtask latent memory weaving for repetitive robot manipulation. arXiv preprint arXiv:2606.17463 (2026),https://arxiv.org/abs/ 2606.17463

2026 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Chen, T., Wang, Y., Li, M., Qin, Y., Shi, H., Li, Z., Hu, Y., Zhang, Y., Wang, K., Chen, Y., Wang, H., Xu, R., Wu, R., Mu, Y., Yang, Y., Dong, H., Luo, P.: Rmbench: Memory-dependent robotic manipulation benchmark with insights into policy design (2026),https://arxiv.org/abs/2603.01229

arXiv 2026

[2] [2]

Chi, C., Xu, Z., Feng, S., Cousineau, E., Du, Y., Burchfiel, B., Tedrake, R., Song, S.: Diffusion policy: Visuomotor policy learning via action diffusion. Int. J. Robotics Res.44(10-11), 1684–1704 (2025).https : / / doi . org / 10 . 1177 / 02783649241273668,https://doi.org/10.1177/02783649241273668

doi:10.1177/02783649241273668 2025

[3] [3]

In: Proceedings of Robotics: Science and Systems

Kim, M.J., Finn, C., Liang, P.: Fine-Tuning Vision-Language-Action Models: Op- timizing Speed and Success. In: Proceedings of Robotics: Science and Systems. LosAngeles, CA, USA (June 2025).https://doi.org/10.15607/RSS.2025.XXI. 017

doi:10.15607/rss.2025.xxi 2025

[4] [4]

In: Agrawal, P., Kroemer, O., Burgard, W

Kim, M.J., Pertsch, K., Karamcheti, S., Xiao, T., Balakrishna, A., Nair, S., Rafailov, R., Foster, E.P., Sanketi, P.R., Vuong, Q., Kollar, T., Burchfiel, B., Tedrake, R., Sadigh, D., Levine, S., Liang, P., Finn, C.: Openvla: An open- source vision-language-action model. In: Agrawal, P., Kroemer, O., Burgard, W. (eds.) Conference on Robot Learning, 6-9 Nov...

work page 2025

[5] [5]

In: The Fourteenth International Conference on Learning Representations (2026),https: //openreview.net/forum?id=KcJ9U0x6kO

Koo, M., Choi, D., Kim, T., Lee, K., Kim, C., Seo, Y., Shin, J.: HAMLET: Switch your vision-language-action model into a history-aware policy. In: The Fourteenth International Conference on Learning Representations (2026),https: //openreview.net/forum?id=KcJ9U0x6kO

work page 2026

[6] [6]

In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S

Liu, B., Zhu, Y., Gao, C., Feng, Y., Liu, Q., Zhu, Y., Stone, P.: Libero: Bench- marking knowledge transfer for lifelong robot learning. In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S. (eds.) Advances in Neu- ral Information Processing Systems. vol. 36, pp. 44776–44791. Curran Associates, Inc. (2023),https://proceedings.neurips.c...

work page 2023

[7] [7]

arXiv preprint arXiv:2602.15010 (2026),https://arxiv.org/abs/2602.15010

Mark, M.S., Liang, J., Attarian, M., Fu, C., Dwibedi, D., Shah, D., Kumar, A.: BPP: Long-context robot imitation learning by focusing on key history frames. arXiv preprint arXiv:2602.15010 (2026),https://arxiv.org/abs/2602.15010

arXiv 2026

[8] [8]

In: Proceedings of Robotics: Science and Systems (2026)

Shah, R., Li, Y., Bello, F., Zhu, Y., Martín-Martín, R.: Memory retrieval in visuo- motor policies for long-horizon robot control. In: Proceedings of Robotics: Science and Systems (2026)

work page 2026