Pith.

Robotics · arXiv:2610.30478 v1 · submitted 21 Oct 2026 · 77 references

Externalise the Memory: A Pre-Registered Study of Where an Embodied Agent's Experience Should Live

Cheng-Yin Tu

The paper says

“Should an embodied agent's experience be baked into its policy weights, or kept external and retrieved at inference?”

more of the abstract

A memory-augmented teacher (a VLM that retrieves lessons, with their episode images, from its own past attempts) produces trajectories on a vision-conditioned reach task; from the same trajectories we distil LoRA student policies and evaluate them under matched protocols, including one that re-attaches the teacher's frozen retrieval buffer to a fixed set of student weights. We pre-registered the hypotheses, arms, tests, and analysis rules before training and before any p-value. Distilling the memory-augmented behaviour into weights beat an action-only control by +1 pp (n.s.); the registered test proves underpowered for its own minimum effect, so we withdraw - rather than refute - the memory-in-weights claim. Attaching retrieval to a fixed set of student weights added +34 pp (z=5.15), reaching the pooled memory-augmented teacher reference (49.3%); the pre-registered protocol contrast agrees (+23 pp, z=3.36); under a pre-registered auto-reset sensitivity count the two are +36 and +29 pp. An exploratory behavioural probe with pre-registered per-arm predictions is consistent with behavioural distillation transferring the marginal correction but not the conditional, situation-specific correction that retrieval supplies. The recipe that follows is to externalise the memory: in follow-up controls registered before they ran, the same frozen buffer lifts the base model with no adapter to 48% (sensitivity 48%) and the other three adapters from 19-26% to 53-55% (sensitivity 17-21% to 50-55%), and the adapter adds nothing detectable once the buffer is attached. On a second, harder task, the retrieval advantage points the same direction but falls short of significance (+6 pp, 95% CI [-4.5, +16.4]); exploratory diagnostics tie the attenuation to memory coverage.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 23 Oct 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

Holds up. Methodologically rigorous empirical study showing that for this embodied task, retrieval vastly outperforms weight consolidation; the proposed marginal/conditional mechanism is a plausible but unconfirmed interpretation. This line is wrong

2 major objections · 6 minor

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 77 references say.

The paper says, at §6.1, second paragraph

The review did not quote this passage.

A classifier read this sentence of the paper as acknowledging the concern (probability 0.78):

“The paper’s claim is scoped more narrowly, pre-registered as aconditional: Amendment 2 pre-specified thata significant, same-direction contrast would constitute cross-task replication of the identical-weights retrieval effect(not of the full 2×2).”

Pith says: Objection 1 of 2 · major

The sentence 'This is the criterion D11 was ultimately built to test, and it is the transferable claim we carry out of a single-recipe study across two tasks' could be read as claiming cross-task validation. However, the L2 contrast (C_retrieval vs A_ctrl_rat, +6 pp, n.s.) did not meet the pre-registered replication criterion (Amendment 2), and the mechanism evidence on L2 is post-hoc with a null σ test.

more of objection 1

The paper's own §5.9.1 and §5.9.3 carefully scope the L2 result to 'mechanism-and-direction reading, not effect-size replication.' To avoid overstatement, please revise the above sentence to explicitly note that the cross-task support is only directional and exploratory, consistent with the scoping already in §5.9.

The paper says, at §4.3, R1∆X variance analysis

The review did not quote this passage.

Pith says: Objection 2 of 2 · major

The variance contrast (σ=14.4 for C_retrieval vs ~9 for distilled arms) is the primary visual evidence for the conditional-correction claim. This analysis was not pre-registered (the probe itself was exploratory per Amendment 14 §14.5) and the paper transparently labels it as such.

more of objection 2

However, given that the mechanistic claim is central to the paper's transferable takeaway, the exploratory label should be kept prominent in the summary and discussion. The current discussion (§6.1) repeats the claim without restating that it rests on exploratory, non-pre-registered evidence. Please add an explicit caveat (e.g., 'This mechanism reading is supported only by exploratory analyses that were not pre-registered for the variance comparison; it requires confirmatory replication.') in §6.1 where the criterion is introduced.

6 minor objections

The paper says, at Abstract

Pith says: minor objection 1 of 6

Line 8: 'the registered test proves underpowered for its own minimum effect' – awkward phrasing. Suggest: 'the registered test is underpowered for its own minimum effect threshold, so we withdraw rather than refute the memory-in-weights claim.'

The paper says, at §3.2

Pith says: minor objection 2 of 6

The sentence 'The recap prompt received an unformatted {target_color} instruction template (disclosed in §4.5)' is ambiguous. Clarify whether the unformatted template appears only in the buffer entries or also in the prompts. The instrument-defect item in §4.5 helps, but consider adding one sentence in §3.2 explaining the effect.

show 4 more

The paper says, at Table 4, arm definitions

Pith says: minor objection 3 of 6

Table 4 uses the term 'D10-ext buffer' for C_retrieval's inference retrieval, but the Methods (§3.4) refer to 'frozen teacher buffer snapshot.' Harmonize the terminology for consistency.

The paper says, at §5.9.3, diagnostic (a)

Pith says: minor objection 4 of 6

The correlation across two tasks (L0a vs L2) is reported as 'Spearman ρ = -0.58' (implied from the text). This is a sample of n=2 tasks, which cannot support a reliable correlation estimate. The paper already says 'n=2 tasks' and calls it exploratory, but the quantitative ρ may give a false impression of precision. Consider presenting only the medians and the effect direction, without claiming a correlation coefficient.

The paper says, at §4.4b, Table 2

Pith says: minor objection 5 of 6

The p-values for the round-3 correlations are permutation-based. State the number of permutations (2000) in the table caption or footnote, as is done in the text. Also note that the Bonferroni-corrected threshold for 12 tests (4 arms × 3 rounds) would be 0.0042; B_main's p=0.001 survives, but A_ctrl_rat's p=0.013 does not. The text does not discuss multiplicity for these post-hoc correlations.

The paper says, at Appendix A (Teacher configuration)

Pith says: minor objection 6 of 6

The asymmetry in recap generation (reasoning-on for L0a vs reasoning-off for L2) is disclosed, but the potential impact on recap quality is not discussed. Since recaps are the content of the conditional correction, a difference in reasoning mode could affect the nature of the retrieved context. Please add a brief note on whether and how this might influence the comparative interpretation.

What would settle it

Re-run the L0a experiment with n=400 episodes per arm to increase power for the mechanism probe; if the R1∆X variance ratio (C_retrieval σ ≈14.4 vs distilled σ ≈9) replicates and the R1' slope remains positive (r>0.2), the conditional-correction interpretation gains independent support. If the variance difference attenuates, retrieval-noise may dominate.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 23 Oct 2026

  1. Reader holds up confidence high
  2. Skeptic agrees 34 s · 31,752 tokens read, 3,274 written
  3. Editor wrote the verdict above 16 s · 32,483 tokens read, 1,526 written
  4. Referee minor revision: 2 major objections, 6 minor 38 s · 31,991 tokens read, 3,732 written

Not run on this paper: the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

The headline result here is real and worth taking seriously: on fixed student weights, adding retrieval boosted success by +34 pp (p≈3×10⁻⁷). The core empirical claim—that for this recipe and task, memory lives in context, not in weights—is well-supported.

What this paper does really well is its methodological discipline. Full pre-registration with amendment chain, honest power analysis on the null, clear separation of confirmatory versus exploratory, and careful scope limitations. The paper sets a standard for how to report a mixed-result study without overclaiming.

The conceptual contribution—the marginal/conditional distinction—is interesting as a hypothesis, but the evidence for it is exploratory: the variance analysis and R1∆X probe were not pre-registered for the mechanism claim, and the cross-task replication (L2) was direction-consistent but non-significant. The authors are upfront about this, but readers should take the mechanism as a plausible post-hoc interpretation, not a confirmed law.

Soft spots: (1) The mechanism claim rests on exploratory analyses with moderate sample sizes. (2) The L2 null limits generality and was predicted by the authors' own power analysis. (3) The distillation null is underpowered, so the claim that weights 'can't hold conditional knowledge' is less certain than the retrieval positive result. The paper acknowledges all of these.

Who is it for? Anyone building memory-augmented embodied agents, and anyone interested in rigorous pre-registered experimentation in ML. It deserves peer review: the empirical core is solid, the interpretation is honestly scoped, and even if the mechanism doesn't generalize, the finding that for this setting retrieval trumps distillation is a contribution. I'd recommend acceptance with minor revision, mainly to keep the exploratory status of the mechanism explicit in the final version.

The strongest objection

The paper's central interpretive claim (marginal knowledge transfers into weights; conditional knowledge belongs in context) rests on the exploratory R1∆X probe and variance analysis, which the paper transparently labels as exploratory and not pre-registered for the specific variance comparison. The R1' slope analysis (§4.2b) partially addresses the concern that retrieval noise could produce the variance; it shows with-buffer corrections track the required correction (r=+0.29 to +0.49), consistent with principled conditional structure rather than pure noise. The L2 σ test was null (F=1.25), but the paper acknowledges the cross-task replication is incomplete and does not overclaim. The reader's weakest_assumption correctly identifies this as the weakest link, but the paper's honesty about the evidence level and the robustness of the confirmatory retrieval effect (+34 pp, z=5.15, p≈2.6e-7) support the ACCEPT verdict.

The reader's weakest_assumption identifies the same concern I have: the marginal/conditional mechanism is supported primarily by exploratory analyses. The paper is transparent about this limitation, scopes its claim accordingly, and does not present the mechanism as proven. The core empirical finding—retrieval on fixed weights adds +34 pp—is backed by confirmatory tests with high significance and consistent sensitivity checks. The L2 result is n.s. but the paper honestly reports it and limits the cross-task generality claim. I see no additional load-bearing concern that would change the verdict. The paper sets a high standard for pre-registration and transparent reporting.

The referee's report

If the results hold, they provide a clean, controlled demonstration that for this distillation recipe (SFT+composable-KTO, LoRA rank-16) on these tasks, the conditional component of retrieved experience does not consolidate into weights, while the marginal component does. The marginal/conditional distinction offers a design principle for where to route different types of knowledge in embodied agents and connects to complementary-learning-systems theory. The paper's strength lies in its rigorous pre-registration, transparent reporting of limitations, and the fact that even the exploratory mechanism probe was pre-registered with per-arm predictions. The confirmatory retrieval effect is robust and survives the pre-registered alpha. The cross-task evidence is weaker but honestly scoped. The work is likely to influence how the community thinks about memory substrates in robotics and VLA post-training.

Pith's reading of the paper (a machine's summary, not the authors')

The paper asks a controlled pre-registered question about an embodied reaching agent: when a memory-augmented teacher produces trajectories, should the student bake that experience into its LoRA weights during distillation, or should it keep the teacher's retrieval buffer external and consult it at inference? The answer is both, but the distinction is which part of the knowledge. Distillation moved the average correction (a static prior) into weights, but the student gained only +1 percentage point over an action-only baseline. By contrast, attaching the same frozen retrieval buffer to a fixed set of student weights added +34 points, matching the teacher's own success rate. An exploratory behavioural probe suggests the mechanism: the student's weights captured the constant marginal shift but not the per-episode conditional adjustment that retrieval supplies. The paper therefore recommends externalising memory for situational corrections and relying on weights only for the average behaviour.

What carries the argument. The central mechanism is the marginal/conditional distinction, operationalised through the R1∆X behavioural probe: the round-1 lateral displacement toward the target. The four distilled arms all cluster at a static prior of about -16.7 grid units (close to the memory-teacher's marginal fingerprint), while retrieval produces a per-episode variable correction with mean near zero and large variance (σ=14.4). This split provides the evidence that distillation transfers the average but retrieval supplies the situation-specific adjustment.

Core claim. The core discovery is a split in what kind of knowledge each substrate holds when an embodied agent learns from a memory-augmented teacher. Under a pre-registered design that isolates the variable, the paper shows that distilling the teacher's trajectories into LoRA weights transfers the marginal correction—the average distribution shift that applies across all episodes—but fails to transfer the conditional, situation-specific correction that varies per episode. When the same retrieval buffer is attached at inference time to identical student weights, the student recovers the full teacher-level success by supplying that conditional correction per-episode. The result is summarised as: 'Marginal knowledge (the average correction that holds across situations) transfers into weights; conditional knowledge (the correction that depends on the situation) belongs in context, because weights approximate it with a constant and context conditions on it for free.'

The premise it rests on. The claim that distillation transfers only the marginal correction while retrieval supplies the conditional correction rests heavily on an exploratory behavioural probe (R1∆X) whose units were corrected via an erratum after the fact, and whose variance analysis (the σ contrast) was not pre-registered; without this probe the mechanism reading would lack direct evidence.

If the paper is right (Pith's inferences, not the authors')

System builders should externalise the memory: keep a retrieval buffer of episodic experience for situation-specific corrections, while using weights only for the average behavioural prior; the adapter adds nothing detectable once the buffer is attached.

The recipe for parametric memory consolidation should target the conditional component, not just the marginal; flat distillation (SFT + KTO) is insufficient for transferring situation-specific corrections.

In the L2 harder task the retrieval effect was directionally consistent but non-significant (+6 pp), and the attenuation is linked to thinner memory coverage and a distance-gated conversion threshold, suggesting the effect scales with the amount of retrievable success experience.

The pre-registered null on parametric memory is scoped to this recipe, data scale, and adapter capacity; stronger methods (surprise-gated writing, multi-round training) remain untested and may capture the conditional structure.

If the marginal/conditional split generalises beyond reaching tasks, then lifelong learning agents should architecturally separate a slow weights channel (for average corrections) from a fast retrieval channel (for situational adjustments), with the retrieval channel indexed by state similarity to the current situation.

The exploratory round-3 correlations (B_main r=+0.40) hint that a conditional trace may exist in later rounds; a round-resolved analysis with more statistical power or a round-by-round distillation target could determine whether the conditional structure is genuinely absent from weights or simply under-learned at round 1.

The buffer-provenance caveat (teacher's buffer vs student's own buffer) suggests that a student retrieving its own self-generated lessons might produce different conditional corrections; this is an open question that could be tested by running a separate arm where the student builds its own buffer through self-exploration.

The paper's pre-registration stance—with a full amendment chain and a locked analysis rule before p-values—offers a methodological template for reporting null results in embodied learning, which may encourage more honest disclosure of non-significant findings in the field.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Should an embodied agent's experience be baked into its policy weights, or kept external and retrieved at inference? A memory-augmented teacher (a VLM that retrieves lessons, with their episode images, from its own past attempts) produces trajectories on a vision-conditioned reach task; from the same trajectories we distil LoRA student policies and evaluate them under matched protocols, including one that re-attaches the teacher's frozen retrieval buffer to a fixed set of student weights. We pre-registered the hypotheses, arms, tests, and analysis rules before training and before any p-value. Distilling the memory-augmented behaviour into weights beat an action-only control by +1 pp (n.s.); the registered test proves underpowered for its own minimum effect, so we withdraw - rather than refute - the memory-in-weights claim. Attaching retrieval to a fixed set of student weights added +34 pp (z=5.15), reaching the pooled memory-augmented teacher reference (49.3%); the pre-registered protocol contrast agrees (+23 pp, z=3.36); under a pre-registered auto-reset sensitivity count the two are +36 and +29 pp. An exploratory behavioural probe with pre-registered per-arm predictions is consistent with behavioural distillation transferring the marginal correction but not the conditional, situation-specific correction that retrieval supplies. The recipe that follows is to externalise the memory: in follow-up controls registered before they ran, the same frozen buffer lifts the base model with no adapter to 48% (sensitivity 48%) and the other three adapters from 19-26% to 53-55% (sensitivity 17-21% to 50-55%), and the adapter adds nothing detectable once the buffer is attached. On a second, harder task, the retrieval advantage points the same direction but falls short of significance (+6 pp, 95% CI [-4.5, +16.4]); exploratory diagnostics tie the attenuation to memory coverage.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-10-23

The authors' main claim, as the classifier picked it out 0.99

“The recipe that follows is to externalise the memory: in follow-up controls registered before they ran, the same frozen buffer lifts the base model with no adapter to 48% (sensitivity 48%) and the other three adapters from 19-26% to 53-55% (sensitivity 17-21% to 50-55%), and the adapter adds nothing detectable once the buffer is attached.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.97 · human participants 0.04 · controlled comparison 0.98 · randomized assignment 0.05 · states quantitative result 0.99 · compares with prior results 0.96 · mentions code or data availability 0.04

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Should an embodied agent's experience be baked into its policy weights, or kept external and retrieved at inference? background 0.77 0.04 0.03 unclear 0.99
A memory-augmented teacher (a VLM that retrieves lessons, with their episode images, from its own past attempts) produces trajectories on a vision-conditioned reach task; from the same trajectories we distil LoRA student policies and evaluate them under matched protocols, including one that re-attaches the teacher's frozen retrieval buffer to a fixed set of student weights. method 0.95 0.94 0.03 asserted 0.86
We pre-registered the hypotheses, arms, tests, and analysis rules before training and before any p-value. method 0.84 0.93 0.02 asserted 1.00
Distilling the memory-augmented behaviour into weights beat an action-only control by +1 pp (n.s.); the registered test proves underpowered for its own minimum effect, so we withdraw - rather than refute - the memory-in-weights claim. result 0.63 0.96 0.07 asserted 0.98
Attaching retrieval to a fixed set of student weights added +34 pp (z=5.15), reaching the pooled memory-augmented teacher reference (49.3%); the pre-registered protocol contrast agrees (+23 pp, z=3.36); under a pre-registered auto-reset sensitivity count the two are +36 and +29 pp. result 1.00 0.96 0.31 asserted 1.00
An exploratory behavioural probe with pre-registered per-arm predictions is consistent with behavioural distillation transferring the marginal correction but not the conditional, situation-specific correction that retrieval supplies. interpretation 0.57 0.94 0.07 hedged possibility 1.00
The recipe that follows is to externalise the memory: in follow-up controls registered before they ran, the same frozen buffer lifts the base model with no adapter to 48% (sensitivity 48%) and the other three adapters from 19-26% to 53-55% (sensitivity 17-21% to 50-55%), and the adapter adds nothing detectable once the buffer is attached. result 0.87 0.96 0.64 asserted 1.00
On a second, harder task, the retrieval advantage points the same direction but falls short of significance (+6 pp, 95% CI [-4.5, +16.4]); exploratory diagnostics tie the attenuation to memory coverage. result 0.91 0.95 0.05 hedged possibility 0.73

What kind of paper the classifier reports this is: empirical study 0.98 · evidence from observation or experiment 0.96

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-10-23: 1 request, 6,807 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (2)
Figure 1 of the paper
Figure 1. Where memory lives decides whether it transfers. Top: one representative episode per arm (final frame; the red cube is the left-arm target). A_ctrl_rat (the same-weights, no-retrieval control) and B_main (the full-teacher-behaviour distilled arm) both time out, while C_retrieval (the same A_ctrl_rat weights with an inference-time retrieval preamble) reaches the target. Over￾laid is the live end-effector-to-target distance; each panel’s success rate (n/100) is the official episode￾level rate, so a single rollout is one representative sample of 100, not the evidence itself. Episode…
Figure 2 of the paper
Figure 2. The R1 mechanism, one panel. Round-1 ∆X toward the target (scalar-guard grid units; 1 unit = 0.45 cm along x), per arm, against two registered reference fingerprints: the D6 memoryless teacher (−23.5) and the memory-teacher last quartile (−15.8). The four distilled arms sit tight on the −15.8 line at ≈−16.7±9, one static prior every arm adopts regardless of episode, while C_retrieval shifts to −7.1±14.4: mean toward zero and a wider, per-episode spread, the signature of a conditional correction. Mean ± SD per arm, plotted from the per-arm summary statistics stored by the original…

Figures are the authors', reproduced from the paper.

References (77)
  1. [1] Wang, Guanzhi and Xie, Yuqi and Jiang, Yunfan and Mandlekar, Ajay and Xiao, Chaowei and Zhu, Yuke and Fan, Linxi and Anandkumar, Anima , journal =. 2024 , note = on Pith
  2. [2] Behrouz, Ali and Zhong, Peilin and Mirrokni, Vahab , booktitle =. 2025 , doi = on Pith
  3. [3] Zhao, Andrew and Huang, Daniel and Xu, Quentin and Lin, Matthieu and Liu, Yong-Jin and Huang, Gao , booktitle =. 2024 , note = on Pith
  4. [4] Hsieh, Cheng-Yu and Li, Chun-Liang and Yeh, Chih-kuan and Nakhost, Hootan and Fujii, Yasuhisa and Ratner, Alex and Krishna, Ranjay and Lee, Chen-Yu and Pfister, Tomas , booktitle =. 2023 , doi = on Pith
  5. [5] Zweiger, Adam and Pari, Jyothish and Guo, Han and Aky. Advances in Neural Information Processing Systems (NeurIPS) , volume =. 2025 , doi = on Pith
  6. [6] Wang, Zora Zhiruo and Mao, Jiayuan and Fried, Daniel and Neubig, Graham , booktitle =. 2025 , note = on Pith
  7. [7] Liu, Zuxin and Zhang, Jesse and Asadi, Kavosh and Liu, Yao and Zhao, Ding and Sabach, Shoham and Fakoor, Rasool , booktitle =. 2024 , note = on Pith
  8. [8] Li, Haozhan and Zuo, Yuxin and Yu, Jiale and Zhang, Yuhao and Yang, Zhaohui and Zhang, Kaiyan and Zhu, Xuekai and Zhang, Yuchen and Chen, Tianxing and Cui, Ganqu and Wang, Dehui and Luo, Dingxiang and Fan, Yuchen and Sun, Youbang and Zeng, Jia and Pang, Jiangmiao and Zhang, Sh... on Pith
  9. [9] Hu, Edward J. and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle =. 2022 , note = on Pith
  10. [10] Ethayarajh, Kawin and Xu, Winnie and Muennighoff, Niklas and Jurafsky, Dan and Kiela, Douwe , booktitle =. 2024 , note = on Pith
  11. [11] Liu, Bo and Zhu, Yifeng and Gao, Chongkai and Feng, Yihao and Liu, Qiang and Zhu, Yuke and Stone, Peter , booktitle =. 2023 , doi = on Pith
  12. [12] Mittal, Mayank and Yu, Calvin and Yu, Qinxi and Liu, Jingzhou and Rudin, Nikita and Hoeller, David and Yuan, Jia Lin and Singh, Ritvik and Guo, Yunrong and Mazhar, Hammad and Mandlekar, Ajay and Babich, Buck and State, Gavriel and Hutter, Marco and Garg, Animesh , journal =. 2... on Pith
  13. [13] Oquab, Maxime and Darcet, Timoth. Transactions on Machine Learning Research , year = on Pith
  14. [14] Lewis, Patrick and Perez, Ethan and Piktus, Aleksandra and Petroni, Fabio and Karpukhin, Vladimir and Goyal, Naman and K. Advances in Neural Information Processing Systems (NeurIPS) , year = on Pith
  15. [15] Zawalski, Micha. Proceedings of The 8th Conference on Robot Learning (CoRL) , series =. 2025 , note = on Pith
  16. [17] Runyu Lu and Yubo Wu and Ethan Kou and Letian Fu and Wenli Xiao and Ajay Mandlekar and Yinzhen Xu and Guanya Shi and Ken Goldberg and Ang Chen and Mosharaf Chowdhury and Yuke Zhu and Linxi Fan and Guanzhi Wang , journal = on Pith
  17. [18] Yuankai Luo and Woping Chen and Tong Liang and Zhenguo Li , journal = on Pith
  18. [19] Brian A. Nosek and Charles R. Ebersole and Alexander C. DeHaven and David T. Mellor , journal =. 2018 , doi = on Pith
  19. [20] Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory , author =. Psychological Review , volume = on Pith
  20. [21] What learning systems do intelligent agents need? Complementary learning systems theory updated , author =. Trends in Cognitive Sciences , volume = on Pith
show 57 more
  1. [22] Ovadia, Oded and Brief, Menachem and Mishaeli, Moshik and Elisha, Oren , booktitle =. 2024 , note = on Pith
  2. [23] Zhu, Yichen and Ou, Zhicai and Mou, Xiaofeng and Tang, Jian , booktitle = on Pith
  3. [24] Memmel, Marius and Berg, Jacob and Chen, Bingqing and Gupta, Abhishek and Francis, Jonathan , booktitle =. 2025 , note = on Pith
  4. [26] Syed, Shahram Najam and Ahuja, Yatharth and Jakobsson, Arthur and Ichnowski, Jeff , booktitle =. 2026 , note = on Pith
  5. [27] Zhao, Jianchao and Yang, Huoren and Hu, Yusong and Gao, Yuyang and Ou, Qiguan and Wan, Cong and Dong, SongLin and Ma, Zhiheng and Gong, Yihong , journal = on Pith
  6. [28] Shi, Hao and Xie, Bin and Liu, Yingfei and Sun, Lin and Liu, Fengrong and Wang, Tiancai and Zhou, Erjin and Fan, Haoqiang and Zhang, Xiangyu and Huang, Gao , booktitle =. 2026 , note = on Pith
  7. [29] Dai, Yinpei and Fu, Hongze and Lee, Jayjun and Liu, Yuejiang and Zhang, Haoran and Yang, Jianing and Finn, Chelsea and Fazeli, Nima and Chai, Joyce , booktitle =. 2026 , note = on Pith
  8. [30] Ren, Tao and Luo, Weiyao and Yang, Hui and Zhu, Rongzhi and Huang, Xiang and Wu, Yuchuan and Chou, Bingxue and Ye, Jieping and Liang, Jiafeng and Li, Yongbin and Peng, Yijie , journal = on Pith
  9. [31] Wu, Rong and Fu, Daocheng and Wen, Licheng and Yang, Xuemeng and Zou, Shu and Mei, Jianbiao and Wang, Yuxin and Zhang, Hairong and Yang, Yu and Hu, Tao and Zhang, Cong and Shi, Botian and Cai, Pinlong , journal = on Pith
  10. [32] Chen, Yuxin and Srikanth, Hari and Jew, Nathan and Wu, Menglin and Wang, Pengcheng and Ren, Junli and Tomizuka, Masayoshi and Xu, Peng and Xie, Jinyu and Tian, Thomas , journal = on Pith
  11. [33] Wu, Beining and Ding, Zihao and Huang, Jun , journal = on Pith
  12. [34] Lu, Haodong and Gong, Dong , journal = on Pith
  13. [35] Xie, Can and Zhou, Yuyi and Yang, Wen and Zhang, Ziyi and Song, Siyao and others , journal =. 2026 , note = on Pith
  14. [37] Jeong, Yuhwan and Yoon, Kuk-Jin , journal =. 2026 , note = on Pith
  15. [38] Ashok, Dhananjay and Agarwal, Shantanu and Datla, Vivek and May, Jonathan and Samuel, Alfy , journal = on Pith
  16. [39] Xia, Siyu and Zhang, Chenheng and Wu, Yanting and Li, Haoxuan and Chai, Jiajun and others , journal = on Pith
  17. [40] Huang, Wei-Chieh and Zhang, Weizhi and Wu, Yuchen and Chen, Yankai and Jiang, Eric Hanchen and others , journal = on Pith
  18. [41] Dhananjay Ashok, Shantanu Agarwal, Vivek Datla, Jonathan May, and Alfy Samuel. How To Train Your World Model: Fine-tuning vs RAG for LM-based World Modeling . arXiv preprint arXiv:2610.02542, 2026 on PithPith's reviewarXiv
  19. [42] Ali Behrouz, Peilin Zhong, and Vahab Mirrokni. Titans: Learning to Memorize at Test Time . In Advances in Neural Information Processing Systems (NeurIPS), volume 38, pages 125925--125962, 2025. doi:10.52202/085713-3786. arXiv:2501.00663 on PithPith's reviewarXiv
  20. [43] Luca Bertinetto, Jo \ a o F. Henriques, Samuel Albanie, Michela Paganini, and G \"u l Varol, editors. NeurIPS 2020 Workshop on Pre-registration in Machine Learning , volume 148, 2021. PMLR on Pith
  21. [44] Yuxin Chen, Hari Srikanth, Nathan Jew, Menglin Wu, Pengcheng Wang, Junli Ren, Masayoshi Tomizuka, Peng Xu, Jinyu Xie, and Thomas Tian. CLIFT: Turning Gemini Robotics On-Device into Humanoid Specialists via Non-Invasive Closed-Loop Iterative Fine-Tuning . arXiv preprint arXiv:2... on PithPith's reviewarXiv
  22. [45] Yinpei Dai, Hongze Fu, Jayjun Lee, Yuejiang Liu, Haoran Zhang, Jianing Yang, Chelsea Finn, Nima Fazeli, and Joyce Chai. RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies . In International Conference on Machine Learning (ICML), 2026. arXiv:2603.04639 on PithPith's reviewarXiv
  23. [46] Kawin Ethayarajh, Winnie Xu, Niklas Muennighoff, Dan Jurafsky, and Douwe Kiela. KTO: Model Alignment as Prospect Theoretic Optimization . In International Conference on Machine Learning (ICML), 2024. arXiv:2402.01306 on PithPith's reviewarXiv
  24. [47] Gemma Team . Gemma: Open Models Based on Gemini Research and Technology . arXiv preprint arXiv:2403.08295, 2024 on PithPith's reviewarXiv
  25. [48] Cheng-Yu Hsieh, Chun-Liang Li, Chih-kuan Yeh, Hootan Nakhost, Yasuhisa Fujii, Alex Ratner, Ranjay Krishna, Chen-Yu Lee, and Tomas Pfister. Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes . In Findings of the Associa... on PithPith's reviewarXiv
  26. [49] Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-Rank Adaptation of Large Language Models . In International Conference on Learning Representations (ICLR), 2022. arXiv:2106.09685 on PithPith's reviewarXiv
  27. [50] Wei-Chieh Huang, Weizhi Zhang, Yuchen Wu, Yankai Chen, Eric Hanchen Jiang, et al. Harness the Memory: A Holistic Evaluation of Memory Substrates in Memory Agents . arXiv preprint arXiv:2608.15008, 2026 on PithPith's reviewarXiv
  28. [51] Yuhwan Jeong and Kuk-Jin Yoon. Reuse Before You Retrieve: Diagnosing Headroom and Complementarity for Test-Time Augmentation of Embodied Multimodal Policies . arXiv preprint arXiv:2608.17484, 2026. Accepted to an ECCV 2026 workshop on PithPith's reviewarXiv
  29. [52] Dharshan Kumaran, Demis Hassabis, and James L. McClelland. What learning systems do intelligent agents need? complementary learning systems theory updated. Trends in Cognitive Sciences, 20 0 (7): 0 512--534, 2016 on Pith
  30. [53] Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K \"u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt \"a schel, Sebastian Riedel, and Douwe Kiela. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks . In Advances ... on PithPith's reviewarXiv
  31. [54] Haozhan Li, Yuxin Zuo, Jiale Yu, Yuhao Zhang, Zhaohui Yang, Kaiyan Zhang, Xuekai Zhu, Yuchen Zhang, Tianxing Chen, Ganqu Cui, Dehui Wang, Dingxiang Luo, Yuchen Fan, Youbang Sun, Jia Zeng, Jiangmiao Pang, Shanghang Zhang, Yu Wang, Yao Mu, Bowen Zhou, and Ning Ding. SimpleVLA-RL... on PithPith's reviewarXiv
  32. [55] Bo Liu, Yifeng Zhu, Chongkai Gao, Yihao Feng, Qiang Liu, Yuke Zhu, and Peter Stone. LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning . In Advances in Neural Information Processing Systems (NeurIPS), volume 36, pages 44776--44791, 2023. doi:10.52202/075280-19... on PithPith's reviewarXiv
  33. [56] Zixuan Liu, Joris K \"o ster, Zizhan Zheng, and Siavash Khajavi. When Does Retrieval Help? A Study of In-Context Adaptation in Vision-Language-Action Models . arXiv preprint arXiv:2610.05492, 2026. NeurIPS 2026 Workshop PTA on PithPith's reviewarXiv
  34. [57] Zuxin Liu, Jesse Zhang, Kavosh Asadi, Yao Liu, Ding Zhao, Shoham Sabach, and Rasool Fakoor. TAIL: Task-specific Adapters for Imitation Learning with Large Pretrained Models . In International Conference on Learning Representations (ICLR), 2024. arXiv:2310.05905 on PithPith's reviewarXiv
  35. [58] Haodong Lu and Dong Gong. ASCENT: Online Test-Time Training of Long-Horizon Agents via Self-Distillation of Verified Experience . arXiv preprint arXiv:2610.05303, 2026 on PithPith's reviewarXiv
  36. [59] Runyu Lu, Yubo Wu, Ethan Kou, Letian Fu, Wenli Xiao, Ajay Mandlekar, Yinzhen Xu, Guanya Shi, Ken Goldberg, Ang Chen, Mosharaf Chowdhury, Yuke Zhu, Linxi Fan, and Guanzhi Wang. ASPIRE: Agentic Skills Discovery for Robotics . arXiv preprint arXiv:2607.00272, 2026 on PithPith's reviewarXiv
  37. [60] Yuankai Luo, Woping Chen, Tong Liang, and Zhenguo Li. CORAL: Scalable Multi-Task Robot Learning via LoRA Experts . arXiv preprint arXiv:2603.09298, 2026 on PitharXiv
  38. [61] James L. McClelland, Bruce L. McNaughton, and Randall C. O'Reilly. Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychological Review, 102 0 (3): 0 419--457, 1995 on Pith
  39. [62] Marius Memmel, Jacob Berg, Bingqing Chen, Abhishek Gupta, and Jonathan Francis. STRAP: Robot Sub-Trajectory Retrieval for Augmented Policy Learning . In International Conference on Learning Representations (ICLR), 2025. arXiv:2412.15182 on PithPith's reviewarXiv
  40. [63] Mayank Mittal, Calvin Yu, Qinxi Yu, Jingzhou Liu, Nikita Rudin, David Hoeller, Jia Lin Yuan, Ritvik Singh, Yunrong Guo, Hammad Mazhar, Ajay Mandlekar, Buck Babich, Gavriel State, Marco Hutter, and Animesh Garg. Orbit: A Unified Simulation Framework for Interactive Robot Learni... on PithPith's reviewarXiv
  41. [64] Brian A. Nosek, Charles R. Ebersole, Alexander C. DeHaven, and David T. Mellor. The preregistration revolution . Proceedings of the National Academy of Sciences, 115 0 (11): 0 2600--2606, 2018. doi:10.1073/pnas.1708274114 on Pithdoi:10.1073/pnas.1708274114
  42. [65] Maxime Oquab, Timoth \'e e Darcet, Th \'e o Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Micha... on PithPith's reviewarXiv
  43. [66] Physical Intelligence , Ali Amin, Raichelle Aniceto, Ashwin Balakrishna, Kevin Black, et al. ^ * _ 0.6 : a VLA That Learns From Experience . arXiv preprint arXiv:2511.14759, 2025 on PithPith's reviewarXiv
  44. [67] Tao Ren, Weiyao Luo, Hui Yang, Rongzhi Zhu, Xiang Huang, Yuchuan Wu, Bingxue Chou, Jieping Ye, Jiafeng Liang, Yongbin Li, and Yijie Peng. Scaling Self-Evolving Agents via Parametric Memory . arXiv preprint arXiv:2606.04536, 2026 on PithPith's reviewarXiv
  45. [68] Hao Shi, Bin Xie, Yingfei Liu, Lin Sun, Fengrong Liu, Tiancai Wang, Erjin Zhou, Haoqiang Fan, Xiangyu Zhang, and Gao Huang. MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation . In International Conference on Learning Representation... on PithPith's reviewarXiv
  46. [69] Shahram Najam Syed, Yatharth Ahuja, Arthur Jakobsson, and Jeff Ichnowski. ExpReS-VLA: Specializing Vision-Language-Action Models Through Experience Replay and Retrieval . In IEEE International Conference on Robotics and Automation (ICRA), 2026. arXiv:2511.06202 on PitharXiv
  47. [70] Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. Voyager: An Open-Ended Embodied Agent with Large Language Models . Transactions on Machine Learning Research, 2024. arXiv:2305.16291 on PithPith's reviewarXiv
  48. [71] Zora Zhiruo Wang, Jiayuan Mao, Daniel Fried, and Graham Neubig. Agent Workflow Memory . In Proceedings of the 42nd International Conference on Machine Learning (ICML), volume 267 of Proceedings of Machine Learning Research, pages 63897--63911, 2025. arXiv:2409.07429 on PithPith's reviewarXiv
  49. [72] Beining Wu, Zihao Ding, and Jun Huang. Not All Experience Belongs in the Weights: Component Routing for Self-Improving GUI Agents . arXiv preprint arXiv:2610.01787, 2026 a on PithPith's reviewarXiv
  50. [73] Rong Wu, Daocheng Fu, Licheng Wen, Xuemeng Yang, Shu Zou, Jianbiao Mei, Yuxin Wang, Hairong Zhang, Yu Yang, Tao Hu, Cong Zhang, Botian Shi, and Pinlong Cai. MemHarness: Memory Is Reconstructed, Not Replayed . arXiv preprint arXiv:2607.28272, 2026 b on PithPith's reviewarXiv
  51. [74] Siyu Xia, Chenheng Zhang, Yanting Wu, Haoxuan Li, Jiajun Chai, et al. UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task Streams . arXiv preprint arXiv:2607.26017, 2026 on PithPith's reviewarXiv
  52. [75] Can Xie, Yuyi Zhou, Wen Yang, Ziyi Zhang, Siyao Song, et al. EDGE: Experience-Distillation for Guided Exploration in Agentic Reinforcement Learning . arXiv preprint arXiv:2608.21946, 2026. Accepted to EMNLP 2026 (Main Conference) on PithPith's reviewarXiv
  53. [76] Micha Zawalski, William Chen, Karl Pertsch, Oier Mees, Chelsea Finn, and Sergey Levine. Robotic Control via Embodied Chain-of-Thought Reasoning . In Proceedings of The 8th Conference on Robot Learning (CoRL), volume 270 of Proceedings of Machine Learning Research, pages 3157--... on PithPith's reviewarXiv
  54. [77] Andrew Zhao, Daniel Huang, Quentin Xu, Matthieu Lin, Yong-Jin Liu, and Gao Huang. ExpeL: LLM Agents Are Experiential Learners . In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2024. arXiv:2308.10144 on PithPith's reviewarXiv
  55. [78] Jianchao Zhao, Huoren Yang, Yusong Hu, Yuyang Gao, Qiguan Ou, Cong Wan, SongLin Dong, Zhiheng Ma, and Yihong Gong. Retrieve-then-Steer: Online Success Memory for Test-Time Adaptation of Generative VLAs . arXiv preprint arXiv:2605.10094, 2026 on PithPith's reviewarXiv
  56. [79] Yichen Zhu, Zhicai Ou, Xiaofeng Mou, and Jian Tang. Retrieval-Augmented Embodied Agents . In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 17985--17995, 2024 on Pith
  57. [80] Adam Zweiger, Jyothish Pari, Han Guo, Ekin Aky \"u rek, Yoon Kim, and Pulkit Agrawal. Self-Adapting Language Models . In Advances in Neural Information Processing Systems (NeurIPS), volume 38, pages 82334--82365, 2025. doi:10.52202/085713-2483. arXiv:2506.10943 on PitharXiv
Cited by 1 Pith paper

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment