Pith. sign in

REVIEW 3 major objections 4 minor 86 references

RoboTTT claims that scaling a robot policy's pretraining context to 8K timesteps yields steady closed-loop performance gains and unlocks one-shot imitation and on-the-fly recovery.

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

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

T0 review · deepseek-v4-flash

2026-08-01 23:38 UTC pith:NYS7BWZ2

load-bearing objection Solid systems paper; the scaling-axis headline is underdetermined by a compute/curriculum confound between the 1K and 8K pretraining runs, and the trial counts are thin. the 3 major comments →

arxiv 2607.15275 v1 pith:NYS7BWZ2 submitted 2026-07-16 cs.RO cs.AIcs.LG

RoboTTT: Context Scaling for Robot Policies

Yunfan Jiang , Yevgen Chebotar , Ruijie Zheng , Fengyuan Hu , Yunhao Ge , Jimmy Wu , Tianyuan Dai , Scott Reed
show 3 more authors
Li Fei-Fei Yuke Zhu Linxi "Jim" Fan
This is my paper
classification cs.RO cs.AIcs.LG
keywords long-context policiestest-time trainingfast weightsrobot foundation modelsvision-language-actionone-shot imitationclosed-loop controlcontext scaling
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish that visuomotor context length is a scaling axis for robot foundation models, not just an architectural detail. The authors build RoboTTT, a policy that compresses up to 8,000 timesteps of history into fast weights — parameters updated by gradient descent during both training and deployment — and they report that closed-loop task completion rises steadily with pretraining context length, with the 8K model scoring 63% higher than the same model trained with 1K context. They also show that long-context conditioning unlocks capabilities short-context policies lack: one-shot imitation from a single human video, on-the-fly recovery from its own mistakes, and robustness to external perturbations. If correct, this gives robot learning a new scaling lever that does not grow inference latency, since the recurrent state is fixed-size. The load-bearing evidence is a small set of real-robot rollouts, so the quantitative claims stand or fall on whether those trial counts are reliable.

Core claim

RoboTTT is a sequence model for robot control built by adding test-time-training (TTT) layers to a pretrained vision-language-action flow-matching policy. Its recurrent state is a set of fast weights: a small MLP updated by gradient descent on an inner self-supervised loss at every timestep, during training and at deployment, so the history is compressed into weight space and later retrieved when producing actions. The training recipe — independent noise levels per action chunk in the sequence (sequence action forcing) plus truncated backpropagation through time — makes training on 8K-timestep sequences feasible under a fixed GPU memory budget. On three real-robot assembly tasks the paper re

What carries the argument

The load-bearing mechanism is the TTT layer with fast weights: a two-layer MLP parameterized by weights updated by gradient descent on an inner loss and then applied to the query, so the model's memory is the weight state itself rather than a fixed-size vector or cached keys and values. Register tokens carry vision-language information across timesteps, a learned tanh gate protects the pretrained behavior early in training, and sequence action forcing with TBPTT lets context grow without memory blow-up. DAgger Distillation is a secondary mechanism: failures update fast weights while the loss is masked to human corrections, distilling the failure-to-correction mapping into the weight state.

Load-bearing premise

The headline comparisons rest on 10–20 rollouts per condition with no reported confidence intervals or multiple seeds, so the observed 57–63% gaps and monotone scaling curve could in principle be sampling noise.

What would settle it

Rerun the main comparisons (8K vs 1K context, and the 128-to-8K scaling series) with at least 50 rollouts per condition and per-configuration confidence intervals, ideally with blinded rubric scoring. If the 8K advantage over 1K falls below the noise floor, or the scaling curve flattens or reverses on a third task, the paper's central claim that pretraining context length yields steady closed-loop gains is falsified.

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

If this is right

  • Pretraining context length can be treated as a scaling axis: longer context translates into higher closed-loop task completion for the same model, with no observed saturation between 128 and 8K timesteps.
  • Long-context conditioning is sufficient to enable one-shot imitation from a single human video of an unseen task configuration, a capability short-context and recurrent-memory baselines lack in this setup.
  • A policy can learn to improve on the fly: distilling failure-to-correction pairs into fast weights yields better recovery than standard DAgger fine-tuning on corrections alone.
  • Because the recurrent state is fixed-size (fast weights), inference cost stays constant in context length, making multi-minute memories practical at 30 Hz control.
  • The scaling benefit is tied to the update rule: a gradient-descent fast model benefits from longer context, while a gated linear recurrent memory does not.

Where Pith is reading between the lines

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

  • Editorial inference: if the scaling trend is real, the gains may persist beyond 8K until the fast-weight MLP's capacity or the meta-learned update dynamics saturate; a natural next test is measuring task completion at 16K and 32K with matched compute.
  • Editorial inference: the failure-as-context principle behind DAgger Distillation could extend to other data sources — such as preference judgments or evaluator critiques — where suboptimal behavior is available alongside corrections, without needing new imitation targets.
  • Editorial inference: the one-shot video-imitation result suggests a route to task specification that does not rely on language; a testable extension is whether conditioning on multiple videos or videos with distractors improves robustness and whether fast weights can be reset between tasks.
  • Editorial inference: if context length is a scaling axis, the compute and training-cost tradeoff becomes central; cheaper TTT training techniques or chunkwise training would determine whether this axis is practical at frontier scale.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper introduces RoboTTT, a Test-Time-Training robot policy built on GR00T N1.7, with fast-weight recurrent states updated by gradient descent at train and test time. The training recipe combines sequence action forcing with truncated backpropagation through time to scale pretraining context to 8K timesteps at fixed inference cost. On three real-robot bimanual assembly tasks, the authors report that RoboTTT-8K outperforms a single-step baseline and a matched Gated DeltaNet recurrent memory baseline, that closed-loop performance rises steadily as pretraining context length grows from 128 to 8K, and that long-context conditioning enables one-shot imitation from human video, on-the-fly recovery via DAgger Distillation, and perturbation robustness.

Significance. If the empirical claims hold, the paper is significant: it would be the first demonstration that pretraining context length is a scaling axis for closed-loop robot manipulation, and it would show that gradient-based fast weights are a practical sequence-memory mechanism for deployed VLA policies. The evaluation has real strengths: held-out Circuit configurations (20 train / 60 test), matched post-training across methods, a GDN recurrent baseline with matched layer placement and parameter count, ablations of action forcing and fast-model expressivity, and unusually detailed rubrics and implementation notes in the appendices. The central scaling claim, however, is currently underdetermined because context length is not varied with matched compute or matched curriculum, and the statistical evidence is thin (10-20 rollouts per condition, no confidence intervals). These issues are fixable and do not undermine the architectural contribution itself, but they block acceptance of the paper as it stands.

major comments (3)
  1. [Fig. 8; §3.4; §A.2] The central claim that 'context length is a new scaling axis' is not isolated from pretraining compute. §A.2 states that per-device batch is 4 (global 64) for context lengths ≤4K and 1 (global 16) for >4K, and §3.4 says context length is 'gradually increased' to the target. Thus the 8K run consumes 16×8192=131K timesteps per optimization step versus 64×1024=64K for the 1K run, a 2× gap, and the curriculum also differs. 'The same model pretrained with 1K-timestep context' is therefore not a controlled comparison. GDN's flat curve weakens a pure-compute explanation, but GDN is a different model class, so it cannot control for an interaction between compute and RoboTTT's meta-learned fast-weight dynamics. Please add a compute-matched control (e.g., a 1K model trained with the same global batch/curriculum and scaled total FLOPs) or report token throughput and a matched-batch schedule for eve
  2. [§4; Fig. 8; Tables 1-3] All headline comparisons rest on 10-20 rollouts per condition with no confidence intervals, per-configuration variance, or multiple seeds. Examples: Gear Bot full successes are 2/10 vs 0/10, one-shot imitation is 6/10 vs 0/10 (Table 2), and Pup Go Car is 9/20 vs 3/20 (Table 1). Some differences may be genuine, but as reported they are not statistically quantified, and the 63%/57% scaling gaps in Fig. 8 use a single average without error bars. Please provide bootstrap confidence intervals over held-out configurations (or over seeds), and report per-configuration outcomes. This is load-bearing for the scaling and one-shot-imitation claims.
  3. [§3.4 vs §4; one-shot imitation] The paper's context-length protocol is internally ambiguous. §3.4 says all models are post-trained on each downstream task at 1K context length, and §4 says sequence models 'use a 1K-timestep context.' Yet one-shot imitation is trained by concatenating a human video and a robot trajectory into a single training sequence, which can easily exceed 1K timesteps for 30 Hz control. Either the one-shot experiments use a longer post-training context (contradicting §3.4) or the effective training context is only 1K (contradicting the '8K context' framing). Please clarify the exact context used for post-training and evaluation in every experiment, and state how 'RoboTTT-8K' should be interpreted if deployment is at a shorter context.
minor comments (4)
  1. [Fig. 8 caption] 'All evaluations in this figure predate the DAgger training used for Pup Go Car in the main results' is ambiguous: does it mean the Pup Go Car component of the figure was run before DAgger was introduced, and if so, why is the DAgger-trained model the one reported in Table 1? Please clarify the relationship between Fig. 8 and the main evaluation.
  2. [Fig. 12] The ablation figure reports only relative improvements without numeric values or error bars. Please add exact values, confidence intervals, and trial counts.
  3. [Table 3 and text] The text says 'the long-context methods react successfully more often' but in the tire condition RoboTTT and GDN both recover 18/20; the claim of an advantage over GDN is supported only by the roof condition. Please soften or qualify.
  4. [§4, baseline description] GR00T N1.7 Hist. is described as using 'one history frame'; the claim that 'history alone does not reliably help' is based on a single additional frame and on Pup Go Car, where the history baseline is worse than no history. This is a very narrow form of history augmentation; please acknowledge the limitation.

Circularity Check

0 steps flagged

No significant circularity: the paper's claims are empirical comparisons against external task rubrics; no fitted parameter is relabeled as a prediction.

full rationale

This is an empirical systems paper, not a derivation chain. The central claim—that scaling pretraining context length from 128 to 8K timesteps yields steady closed-loop gains (Sec. 4, Fig. 8)—rests on comparing the same RoboTTT architecture pretrained at different context lengths and post-trained on identical task data with rubric-based completion scores. No model parameter is fitted to the reported completion percentages and then renamed as a prediction; the 8K-vs-1K comparison is an internal experimental contrast, not an identity by construction. The TTT update/apply equations (Eqs. 1–2), the sequence flow-matching loss (Eq. 5), tanh gating (Eq. 3), and TBPTT are stated model mechanics rather than conclusions derived from the benchmarks. Self-citations to GR00T N1.7 and Egoscale are prior-art backbone and pretraining-data usage; these are externally released resources and do not load-bear the scaling conclusion. The one-shot imitation and DAgger Distillation results are trained capabilities evaluated on held-out trials, and while their generalization could be debated, they do not reduce to the training inputs by definition. The main weaknesses—confounding context length with pretraining compute/batch schedule and small trial counts without confidence intervals—are correctness and experimental-validity concerns, not circularity. Accordingly, no circular step can be exhibited, and the score is 0.

Axiom & Free-Parameter Ledger

6 free parameters · 7 axioms · 0 invented entities

The ledger lists hand-chosen hyperparameters and domain assumptions rather than fitted physical constants; the paper is empirical, so most unknown quantities are experimental. No new physical entities or conservation laws are introduced; 'fast weights' and 'register tokens' are adopted from prior work.

free parameters (6)
  • Inner fast-weight learning rate η = learned (base 0.1)
    Controls the TTT update in Eq. 1; trained on top of a hand-set base value (App. A.1); final value not reported.
  • Tanh gate initialization α = 0.001
    Hand-chosen to keep TTT contribution near zero at start of training (Eq. 3).
  • Register token count N = 16
    Hand-chosen; registers are the cross-time carriers of VL information through TTT layers (Sec. 3.1).
  • Fast model capacity = two-layer MLP, ~10M params per TTT layer
    Architecture choice; a linear fast model is 27% worse in ablations (Sec. 4).
  • Sequence action forcing noise schedule = τ=s(1-u), u~Beta(1.5,1), s=0.999
    Flow-matching noise schedule sampled per action chunk (Eq. 5); chosen from prior work.
  • Pretraining/post-training context lengths = up to 8K / 1K timesteps
    The 8K-vs-1K comparison is the scaling claim; post-training is always at 1K (Sec. 3.4, Fig. 8).
axioms (7)
  • domain assumption The real-robot evaluation protocol (fixed initial placements, 4 cameras, 30Hz control) yields measurements that transfer to claims about robot policies in general.
    Sec. 4 and App. B describe the YAM setup; no independent hardware replication is provided.
  • domain assumption Rubric-based task completion scores are valid partial-credit measures of assembly progress.
    App. B.1 defines rubrics; biases in rubric design could favor methods that recover from mistakes.
  • domain assumption The pretraining mixture (tabletop bimanual robot + egocentric human data) is representative enough for TTT fast-weight dynamics learned there to transfer to the three downstream tasks.
    Sec. 3.4/App. A.2; the mixture is curated for long trajectories but not released.
  • domain assumption Fast-weight updates by gradient descent remain stable and informative across 8K timesteps during deployment.
    Central to the method; the paper does not analyze drift, catastrophic forgetting, or update statistics.
  • domain assumption GDN is a fair, sufficiently strong recurrent-memory baseline; RoboTTT-vs-GDN differences are attributed to the update rule.
    Sec. 4: TTT layers replaced with Gated DeltaNet layers, layer placement/gating/param count matched, but tuning effort for GDN is not reported.
  • domain assumption The 10-20 rollouts per condition are enough support for the reported success counts and scaling curve.
    Tables 1-3 and Fig. 8 contain no confidence intervals or seeds.
  • domain assumption In the one-shot imitation experiment, the human video is the only cue identifying the target configuration.
    Sec. 4: prompt is identical across configurations; initial-layout variation is controlled but not quantified.

pith-pipeline@v1.3.0-alltime-deepseek · 21782 in / 15613 out tokens · 142475 ms · 2026-08-01T23:38:42.235677+00:00 · methodology

0 comments
read the original abstract

Recent robot foundation models operate with single-step or short-history visuomotor context. We introduce Test-Time-Training Robot Policies (RoboTTT), a robot model and training recipe that scale visuomotor context to 8K timesteps, three orders of magnitude beyond state-of-the-art policies, without growing inference latency. At this context length, we unlock new robot capabilities: one-shot in-context imitation from human video demonstrations, on-the-fly policy improvement, robustness to perturbations, and stronger performance on multi-stage, long-horizon tasks. We also observe, for the first time, steady gains in closed-loop performance as pretraining context length scales. At its core, RoboTTT integrates Test-Time Training into robot foundation models such as Vision-Language-Action policies, yielding a sequence model whose recurrent state consists of fast weights, parameters updated by gradient descent during both training and inference, compressing histories into weight space and retrieving contextual information for long-context conditioning. To scale training context length, the recipe combines sequence action forcing with truncated backpropagation through time. On challenging real-robot manipulation tasks, RoboTTT improves overall performance by 87% over the single-step context baseline and fully completes a five-minute, ten-stage assembly task, which no baseline ever does. RoboTTT trained with 8K-timestep context outperforms the same model pretrained with 1K timesteps by 62%, suggesting context length as a new scaling axis for robot foundation models. Videos are available at https://research.nvidia.com/labs/gear/robottt/

Figures

Figures reproduced from arXiv: 2607.15275 by Fengyuan Hu, Jimmy Wu, Li Fei-Fei, Linxi "Jim" Fan, Ruijie Zheng, Scott Reed, Tianyuan Dai, Yevgen Chebotar, Yuke Zhu, Yunfan Jiang, Yunhao Ge.

Figure 1
Figure 1. Figure 1: RoboTTT, a long-context visuomotor policy that integrates Test-Time Training (TTT) into robot foundation models, with context scaled to 8K timesteps. RoboTTT exhibits capabilities such as one-shot in-context imitation from human videos and on-the-fly policy improvement. © 2026 NVIDIA. All rights reserved. arXiv:2607.15275v1 [cs.RO] 16 Jul 2026 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: RoboTTT model architecture, training, and inference. TTT layers are added after the attention layers in the DiT action head: attention operates within each timestep, while TTT layers operate across timesteps. Training uses a sequence flow-matching loss with sequence action forcing, sampling the noise level independently per action chunk. Inference starts from the learned initialization 𝑊0, updating fast we… view at source ↗
Figure 3
Figure 3. Figure 3: Computation flow with tanh gat￾ing. The TTT output is weighted by the learned gate tanh(𝛼) before being added to the attention output. Gating for Preserving Pretrained Capabilities To retain the knowledge of the pretrained VLA model, RoboTTT is initialized from the base model weights and adopts a learned tanh gating mechanism (1) that keeps the contribution of TTT small at the start of training. Concretely… view at source ↗
Figure 4
Figure 4. Figure 4: TBPTT. Gradients are truncated at segment boundaries; fast weights carry over, so TTT continues over the entire sequence. Truncated Backpropagation Through Time Training on long sequences with full backpropagation through time (BPTT) stores activations for every timestep, so GPU memory grows with sequence length. We instead adopt truncated backprop￾agation through time (TBPTT): the input sequence is divide… view at source ↗
Figure 5
Figure 5. Figure 5: Evaluation tasks. Three long-horizon assembly tasks on a YAM bimanual setup; each row shows one rollout. Top, Pup Go Car: toy vehicle assembly, 2-minute average episodes. Middle, Circuit: target configuration specified by a language prompt or a one-shot human video demonstration; 1-minute episodes. Bottom, Gear Bot: ten-stage assembly spanning five minutes, our longest-horizon task. 3.3. Effective Learning… view at source ↗
Figure 6
Figure 6. Figure 6: DAgger Distillation. During se￾quence training, all executed actions update the fast weights, but the flow-matching loss is computed only on human corrections. DAgger Distillation Consider a robot rollout with human corrections collected as in DAgger (57): whenever the robot makes a mistake, a human operator intervenes with corrective actions, yielding a trajectory 𝜉 DAgger = {(𝑙, 𝑜𝑡, 𝑞𝑡, 𝐴𝑡)} 𝑇 𝑡=1 in whi… view at source ↗
Figure 7
Figure 7. Figure 7: Main evaluation: task completion scores on three assembly tasks. Scores are rubric-based, reported in percent; higher is better. the improvement process induced by human interventions is distilled into the policy’s fast-weight adaptation. At test time, the model performs such corrections online, without human intervention; its own corrections then enter the history and are absorbed into the fast weights, e… view at source ↗
Figure 8
Figure 8. Figure 8: Closed-loop performance scales with pretraining context length. Average task completion score across the three tasks as pretraining context length grows from 128 to 8K timesteps. RoboTTT improves steadily, surpassing the best short-context baseline from 1K onward with no sign of saturation; GDN does not benefit from longer context. Single-Step Context and Short Context denote GR00T N1.7 and GR00T N1.7 Hist… view at source ↗
Figure 9
Figure 9. Figure 9: One-shot imitation from an in-context human video. Continuous footage: a human demonstrates an unseen configuration (frames 1–3), the scene is reset (frame 4), and RoboTTT reproduces the assembly (row 2). The prompt is identical across configurations, so the target is identifiable only from the video. steadily with pretraining context length, reaching 71.5% at 8K, 63% higher than the same model pretrained … view at source ↗
Figure 10
Figure 10. Figure 10: DAgger Distillation results on Pup Go Car. Task completion after fine-tuning on a pool of 100 DAgger trajectories (50 collected with RoboTTT, 50 with GR00T N1.7). DAgger Distillation applies to the sequence models RoboTTT and GDN. Drill Progress (%) 100 50 0 Time [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: On-the-fly recovery learned through DAgger Distillation. While tightening the roof screw on Pup Go Car, RoboTTT misses the screw (frames 1–3), raises the arm and re-attempts, getting closer but missing again (frames 4–5), then re-adjusts once more and succeeds (frames 6–8). cessfully more often: RoboTTT recovers in 15/20 roof-perturbation trials versus 13/20 for GDN and at most 10/20 for the short-context… view at source ↗
Figure 12
Figure 12. Figure 12: Ablation results on the Pup Go Car task. We ablate sequence action forcing and the fast-model architecture (MLP vs. linear), and trace each component added during development (state tokens, action tokens, register tokens). GR00T N1.7 with register tokens is included for a matched-token comparison. Design choices that matter for RoboTTT’s per￾formance. We ablate key design choices. First, we compare agains… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

86 extracted references · 43 linked inside Pith

  1. [1]

    Flamingo: a visual language model for few-shot learning.arXiv preprint arXiv: 2204.14198, 2022

    Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Mensch, Katie Millican, Malcolm Reynolds, Roman Ring, Eliza Rutherford, Serkan Cabi, Tengda Han, Zhitao Gong, Sina Samangooei, Marianne Monteiro, Jacob Menick, Sebastian Borgeaud, Andrew Brock, Aida Nematzadeh, Sahand Sharifzadeh, Mikolaj Binkowski,...

    Pith/arXiv arXiv 2022

  2. [2]

    Evolve-vla: Test-time training from environment feedback for vision-language-action models.arXiv preprint arXiv: 2512.14666, 2025

    Zechen Bai, Chen Gao, and Mike Zheng Shou. Evolve-vla: Test-time training from environment feedback for vision-language-action models.arXiv preprint arXiv: 2512.14666, 2025. 12

    arXiv 2025

  3. [3]

    Titans: Learning to memorize at test time.arXiv preprint arXiv: 2501.00663, 2024

    Ali Behrouz, Peilin Zhong, and Vahab Mirrokni. Titans: Learning to memorize at test time.arXiv preprint arXiv: 2501.00663, 2024. 12

    Pith/arXiv arXiv 2024

  4. [4]

    It’s all connected: A journey through test-time memorization, attentional bias, retention, and online optimization.arXiv preprint arXiv: 2504.13173, 2025

    Ali Behrouz, Meisam Razaviyayn, Peilin Zhong, and Vahab Mirrokni. It’s all connected: A journey through test-time memorization, attentional bias, retention, and online optimization.arXiv preprint arXiv: 2504.13173, 2025. 12

    Pith/arXiv arXiv 2025

  5. [5]

    org/abs/2410.24164, 2024

    Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, et al.𝜋0: A vision-language-action flow model for general robot control, 2024.URL https://arxiv. org/abs/2410.24164, 2024. 2, 11, 12

    Pith/arXiv arXiv 2024

  6. [6]

    Lee, Maria Bauzá Villalonga, Todor Davchev, Yuxiang Zhou, Agrim Gupta, A

    Konstantinos Bousmalis, Giulia Vezzani, Dushyant Rao, Coline Devin, Alex X. Lee, Maria Bauzá Villalonga, Todor Davchev, Yuxiang Zhou, Agrim Gupta, A. Raju, Antoine Laurens, Claudio Fantacci, Valentin Dalibard, Martina Zambelli, M. F. Martins, Rugile Pevceviciute, M. Blokzijl, Misha Denil, Nathan Batchelor, Thomas Lampe, Emilio Parisotto, Konrad Zolna, Sco...

  7. [7]

    Irpan, Tomas Jackson, Sally Jesmonth, Nikhil J

    AnthonyBrohan, NoahBrown, JusticeCarbajal, YevgenChebotar, JosephDabis, ChelseaFinn, K.Gopalakr- ishnan, Karol Hausman, Alexander Herzog, Jasmine Hsu, Julian Ibarz, Brian Ichter, A. Irpan, Tomas Jackson, Sally Jesmonth, Nikhil J. Joshi, Ryan C. Julian, Dmitry Kalashnikov, Yuheng Kuang, Isabel Leal, Kuang-Huei Lee, S. Levine, Yao Lu, U. Malla, D. Manjunath...

    Pith/arXiv arXiv doi:10.48550/arxiv.2212.06817 2022

  8. [8]

    Choromanski, Tianli Ding, Danny Driess, Kumar Avinava Dubey, Chelsea Finn, Peter R

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, K. Choromanski, Tianli Ding, Danny Driess, Kumar Avinava Dubey, Chelsea Finn, Peter R. Florence, Chuyuan Fu, Montse Gonzalez Arenas, K. Gopalakrishnan, Kehang Han, Karol Hausman, Alexander Herzog, Jasmine Hsu, Brian Ichter, A. Irpan, 13 RoboTTT: Context Scaling for Robot Policies Nikhil J. Jos...

    Pith/arXiv arXiv doi:10.48550/arxiv.2307.15818 2023

  9. [9]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan,PranavShyam,GirishSastry,AmandaAskell,SandhiniAgarwal,ArielHerbert-Voss,Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gra...

    Pith/arXiv arXiv 2005

  10. [10]

    Diffusion forcing: Next-token prediction meets full-sequence diffusion.arXiv preprint arXiv: 2407.01392,

    Boyuan Chen, Diego Marti Monso, Yilun Du, Max Simchowitz, Russ Tedrake, and Vincent Sitzmann. Diffusion forcing: Next-token prediction meets full-sequence diffusion.arXiv preprint arXiv: 2407.01392,

    Pith/arXiv arXiv

  11. [11]

    Ttt3r: 3d reconstruction as test-time training, 2026

    Xingyu Chen, Yue Chen, Yuliang Xiu, Andreas Geiger, and Anpei Chen. Ttt3r: 3d reconstruction as test-time training, 2026. URLhttps://arxiv.org/abs/2509.26645. 12

    Pith/arXiv arXiv 2026

  12. [12]

    Robomme: Benchmarking and understanding memory for robotic generalist policies

    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. arXiv preprint arXiv:2603.04639, 2026. 2

    Pith/arXiv arXiv 2026

  13. [13]

    One-minute video generation with test-time training

    Karan Dalal, Daniel Koceja, Jiarui Xu, Yue Zhao, Shihao Han, Ka Chun Cheung, Jan Kautz, Yejin Choi, Yu Sun, and Xiaolong Wang. One-minute video generation with test-time training. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 17702–17711, June 2025. 12

    2025

  14. [14]

    Vision transformers need registers

    Timothée Darcet, Maxime Oquab, Julien Mairal, and Piotr Bojanowski. Vision transformers need registers. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11,

    2024

  15. [15]

    Causal confusion in imitation learning

    Pim de Haan, Dinesh Jayaraman, and Sergey Levine. Causal confusion in imitation learning. In H. Wal- lach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URLhttps://proceedings. neurips.cc/paper_files/paper/2019/file/947018640bf36a2...

    arXiv 2019

  16. [16]

    Longrope: Extending LLM context window beyond 2 million tokens

    Yiran Ding, Li Lyna Zhang, Chengruidong Zhang, Yuanyuan Xu, Ning Shang, Jiahang Xu, Fan Yang, and Mao Yang. Longrope: Extending LLM context window beyond 2 million tokens. In Ruslan Salakhutdinov, Zico Kolter, Katherine A. Heller, Adrian Weller, Nuria Oliver, Jonathan Scarlett, and Felix Berkenkamp, editors,Forty-first International Conference on Machine ...

    2024

  17. [17]

    Knowledge insulating vision-language-action models: Train fast, run fast, generalize better.Advances in Neural Information Processing Systems, 38:102867–102888,

    Danny Driess, Jost Springenberg, Brian Ichter, Lili Yu, Adrian Li-Bell, Karl Pertsch, Allen Ren, Homer Walke, Quan Vuong, Lucy Xiaoyang Shi, et al. Knowledge insulating vision-language-action models: Train fast, run fast, generalize better.Advances in Neural Information Processing Systems, 38:102867–102888,

  18. [18]

    One-shot imitation learning.Advances in neural information processing systems, 30, 2017

    Yan Duan, Marcin Andrychowicz, Bradly Stadie, OpenAI Jonathan Ho, Jonas Schneider, Ilya Sutskever, Pieter Abbeel, and Wojciech Zaremba. One-shot imitation learning.Advances in neural information processing systems, 30, 2017. 2, 11 14 RoboTTT: Context Scaling for Robot Policies

    2017

  19. [19]

    Molmoact2: Action reasoning models for real-world deployment, 2026

    Haoquan Fang, Jiafei Duan, Donovan Clay, Sam Wang, Shuo Liu, Weikai Huang, Xiang Fan, Wei-Chuan Tsai, Shirui Chen, Yi Ru Wang, Shanli Xing, Jaemin Cho, Jae Sung Park, Ainaz Eftekhar, Peter Sushko, Karen Farley, Angad Wadhwa, Cole Harrison, Winson Han, Ying-Chun Lee, Eli VanderBilt, Rose Hendrix, Suveen Ellawela, Lucas Ngoo, Joyce Chai, Zhongzheng Ren, Ali...

    Pith/arXiv arXiv 2026

  20. [20]

    In-place test-time training

    Guhao Feng, Shengjie Luo, Kai Hua, Ge Zhang, Wenhao Huang, Di He, and Tianle Cai. In-place test-time training. InThe Fourteenth International Conference on Learning Representations, 2026. URL https://openreview.net/forum?id=dTWfCLSoyl. 12

    2026

  21. [21]

    Model-agnostic meta-learning for fast adaptation of deep networks.arXiv preprint arXiv: 1703.03400, 2017

    Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks.arXiv preprint arXiv: 1703.03400, 2017. 5

    Pith/arXiv arXiv 2017

  22. [22]

    In-context imitation learning via next-token prediction.arXiv preprint arXiv:2408.15980, 2024

    Letian Fu, Huang Huang, Gaurav Datta, Lawrence Yunliang Chen, William Chung-Ho Panitch, Fangchen Liu, Hui Li, and Ken Goldberg. In-context imitation learning via next-token prediction.arXiv preprint arXiv:2408.15980, 2024. 3, 11

    Pith/arXiv arXiv 2024

  23. [23]

    Gated memory policy.arXiv preprint arXiv: 2604.18933, 2026

    Yihuai Gao, Jinyun Liu, Shuang Li, and Shuran Song. Gated memory policy.arXiv preprint arXiv: 2604.18933, 2026. 11, 12

    Pith/arXiv arXiv 2026

  24. [24]

    Vit3: Unlocking test-time training in vision.arXiv preprint arXiv: 2512.01643, 2025

    Dongchen Han, Yining Li, Tianyu Li, Zixuan Cao, Ziming Wang, Jun Song, Yu Cheng, Bo Zheng, and Gao Huang. Vit3: Unlocking test-time training in vision.arXiv preprint arXiv: 2512.01643, 2025. 12

    Pith/arXiv arXiv 2025

  25. [25]

    Gaussian error linear units (gelus).arXiv preprint arXiv: 1606.08415,

    Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus).arXiv preprint arXiv: 1606.08415,

    Pith/arXiv arXiv

  26. [26]

    Long short-term memory.Neural Comput., 9(8):1735–1780, November 1997

    Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory.Neural Comput., 9(8):1735–1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URLhttps://doi.org/10.1162/ neco.1997.9.8.1735. 11

    doi:10.1162/neco.1997.9.8.1735 1997

  27. [27]

    Ruler: What’s the real context size of your long-context language models?arXiv preprint arXiv: 2404.06654, 2024

    Cheng-Ping Hsieh, Simeng Sun, Samuel Kriman, Shantanu Acharya, Dima Rekesh, Fei Jia, Yang Zhang, and Boris Ginsburg. Ruler: What’s the real context size of your long-context language models?arXiv preprint arXiv: 2404.06654, 2024. 2

    Pith/arXiv arXiv 2024

  28. [28]

    Minicpm: Unveiling the potential of small language models with scalable training strategies.arXiv preprint arXiv: 2404.06395, 2024

    Shengding Hu, Yuge Tu, Xu Han, Chaoqun He, Ganqu Cui, Xiang Long, Zhi Zheng, Yewei Fang, Yuxiang Huang,WeilinZhao,XinrongZhang,ZhengLengThai,KaihuoZhang,ChongyiWang,YuanYao,Chenyang Zhao, Jie Zhou, Jie Cai, Zhongwu Zhai, Ning Ding, Chao Jia, Guoyang Zeng, Dahai Li, Zhiyuan Liu, and Maosong Sun. Minicpm: Unveiling the potential of small language models wit...

    Pith/arXiv arXiv 2024

  29. [29]

    Physical Intelligence, Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Manuel Y. Galliker, Dibya Ghosh, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Devin LeBlanc, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch...

    Pith/arXiv arXiv 2025

  30. [30]

    Perceiver: General perception with iterative attention.arXiv preprint arXiv: 2103.03206, 2021

    Andrew Jaegle, Felix Gimeno, Andrew Brock, Andrew Zisserman, Oriol Vinyals, and Joao Carreira. Perceiver: General perception with iterative attention.arXiv preprint arXiv: 2103.03206, 2021. 4

    Pith/arXiv arXiv 2021

  31. [31]

    Contextvla: Vision-language-action model with amortized multi-frame context.arXiv preprint arXiv: 2510.04246,

    Huiwon Jang, Sihyun Yu, Heeseung Kwon, Hojin Jeon, Younggyo Seo, and Jinwoo Shin. Contextvla: Vision-language-action model with amortized multi-frame context.arXiv preprint arXiv: 2510.04246,

    arXiv

  32. [32]

    Vima: General robot manipulation with multimodal prompts.arXiv preprint arXiv: 2210.03094, 2022

    Yunfan Jiang, Agrim Gupta, Zichen Zhang, Guanzhi Wang, Yongqiang Dou, Yanjun Chen, Li Fei-Fei, Anima Anandkumar, Yuke Zhu, and Linxi Fan. Vima: General robot manipulation with multimodal prompts.arXiv preprint arXiv: 2210.03094, 2022. 3, 11, 12

    Pith/arXiv arXiv 2022

  33. [33]

    Muon: An optimizer for hidden layers in neural networks, 2024

    Keller Jordan, Yuchen Jin, Vlado Boza, Jiacheng You, Franz Cesista, Laker Newhouse, and Jeremy Bernstein. Muon: An optimizer for hidden layers in neural networks, 2024. URLhttps://kellerjordan. github.io/posts/muon/. 20

    2024

  34. [34]

    Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models.arXiv preprint arXiv: 2001.08361, 2020. 11

    Pith/arXiv arXiv 2001

  35. [35]

    Lattice: Learning to efficiently compress the memory.arXiv preprint arXiv: 2504.05646, 2025

    Mahdi Karami, Razvan Pascanu, and Vahab Mirrokni. Lattice: Learning to efficiently compress the memory.arXiv preprint arXiv: 2504.05646, 2025. 12

    arXiv 2025

  36. [36]

    OpenVLA: An open-source vision-language- action model

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan P Foster, Pannag R Sanketi, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. OpenVLA: An open-source vision-language- action model. In8th Annual Conference on Robot Learni...

    2024

  37. [37]

    Cosmos policy: Fine-tuning video models for visuomotor control and planning.arXiv preprint arXiv: 2601.16163, 2026

    Moo Jin Kim, Yihuai Gao, Tsung-Yi Lin, Yen-Chen Lin, Yunhao Ge, Grace Lam, Percy Liang, Shuran Song, Ming-Yu Liu, Chelsea Finn, and Jinwei Gu. Cosmos policy: Fine-tuning video models for visuomotor control and planning.arXiv preprint arXiv: 2601.16163, 2026. 2, 11

    Pith/arXiv arXiv 2026

  38. [38]

    Rma: Rapid motor adaptation for legged robots.arXiv preprint arXiv:2107.04034, 2021

    Ashish Kumar, Zipeng Fu, Deepak Pathak, and Jitendra Malik. Rma: Rapid motor adaptation for legged robots.arXiv preprint arXiv:2107.04034, 2021. 11

    Pith/arXiv arXiv 2021

  39. [39]

    In-context reinforcement learning with algorithm distillation

    Michael Laskin, Luyu Wang, Junhyuk Oh, Emilio Parisotto, Stephen Spencer, Richie Steigerwald, DJ Strouse, Steven Stenberg Hansen, Angelos Filos, Ethan Brooks, Maxime Gazeau, Himanshu Sahni, Satinder Singh, and Volodymyr Mnih. In-context reinforcement learning with algorithm distillation. In The Eleventh International Conference on Learning Representations...

    2023

  40. [40]

    Causal world modeling for robot control.arXiv preprint arXiv: 2601.21998, 2026

    Lin Li, Qihang Zhang, Yiming Luo, Shuai Yang, Ruilin Wang, Fei Han, Mingrui Yu, Zelin Gao, Nan Xue, Xing Zhu, Yujun Shen, and Yinghao Xu. Causal world modeling for robot control.arXiv preprint arXiv: 2601.21998, 2026. 11

    Pith/arXiv arXiv 2026

  41. [41]

    Unified video action model.arXiv preprint arXiv: 2503.00200, 2025

    Shuang Li, Yihuai Gao, Dorsa Sadigh, and Shuran Song. Unified video action model.arXiv preprint arXiv: 2503.00200, 2025. 2, 11

    Pith/arXiv arXiv 2025

  42. [42]

    Tnt: Improving chunkwise training for test-time memorization.arXiv preprint arXiv:2511.07343, 2025

    Zeman Li, Ali Behrouz, Yuan Deng, Peilin Zhong, Praneeth Kacham, Mahdi Karami, Meisam Razaviyayn, and Vahab Mirrokni. Tnt: Improving chunkwise training for test-time memorization.arXiv preprint arXiv:2511.07343, 2025. 12

    arXiv 2025

  43. [43]

    Parallelizing non-linear sequential models over the sequence length

    Yi Heng Lim, Qi Zhu, Joshua Selfridge, and Muhammad Firmansyah Kasim. Parallelizing non-linear sequential models over the sequence length. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024. URLhttps: //openreview.net/forum?id=E34AlVLN0v. 12

    2024

  44. [44]

    Onetwovla: A unified vision-language-action model with adaptive reasoning.arXiv preprint arXiv:2505.11917, 2025

    Fanqi Lin, Ruiqian Nai, Yingdong Hu, Jiacheng You, Junming Zhao, and Yang Gao. Onetwovla: A unified vision-language-action model with adaptive reasoning.arXiv preprint arXiv:2505.11917, 2025. 11

    arXiv 2025

  45. [45]

    On-the-fly vla adaptation via test-time reinforcement learning.arXiv preprint arXiv:2601.06748, 2026

    Changyu Liu, Yiyang Liu, Taowen Wang, Qiao Zhuang, James Chenhao Liang, Wenhao Yang, Renjing Xu, Qifan Wang, Dongfang Liu, and Cheng Han. On-the-fly vla adaptation via test-time reinforcement learning.arXiv preprint arXiv:2601.06748, 2026. 12 16 RoboTTT: Context Scaling for Robot Policies

    Pith/arXiv arXiv 2026

  46. [46]

    Lost in the middle: How language models use long contexts.Transactions of the association for computational linguistics, 12:157–173, 2024

    Nelson F Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. Lost in the middle: How language models use long contexts.Transactions of the association for computational linguistics, 12:157–173, 2024. 2

    2024

  47. [47]

    RDT-1B:a diffusionfoundationmodelforbimanual manipulation

    Songming Liu, Lingxuan Wu, Bangguo Li, Hengkai Tan, Huayu Chen, Zhengyi Wang, Ke Xu, Hang Su, and JunZhu. RDT-1B:a diffusionfoundationmodelforbimanual manipulation. InTheThirteenthInternational Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025. OpenReview.net, 2025. URLhttps://openreview.net/forum?id=yAzN4tz7oI. 2, 11

    2025

  48. [48]

    Loshchilov and F

    I. Loshchilov and F. Hutter. Decoupled weight decay regularization.International Conference on Learning Representations, 2017. 20

    2017

  49. [49]

    Savarese, Yuke Zhu, and Roberto Mart’in-Mart’in

    Ajay Mandlekar, Danfei Xu, Josiah Wong, Soroush Nasiriany, Chen Wang, Rohun Kulkarni, Li Fei-Fei, S. Savarese, Yuke Zhu, and Roberto Mart’in-Mart’in. What matters in learning from offline human demonstrations for robot manipulation.Conference on Robot Learning, 2021. 11

    2021

  50. [50]

    Bpp: Long-context robot imitation learning by focusing on key history frames.arXiv preprint arXiv: 2602.15010, 2026

    Max Sobol Mark, Jacky Liang, Maria Attarian, Chuyuan Fu, Debidatta Dwibedi, Dhruv Shah, and Aviral Kumar. Bpp: Long-context robot imitation learning by focusing on key history frames.arXiv preprint arXiv: 2602.15010, 2026. 2, 11, 12

    arXiv 2026

  51. [51]

    Gr00t n1: An open foundation model for generalist humanoid robots.arXiv preprint arXiv: 2503.14734, 2025

    NVIDIA, Johan Bjorck, Fernando Castañeda, Nikita Cherniadev, Xingye Da, Runyu Ding, Linxi "Jim" Fan, Yu Fang, Dieter Fox, Fengyuan Hu, Spencer Huang, Joel Jang, Zhenyu Jiang, Jan Kautz, Kaushil Kundalia, Lawrence Lao, Zhiqi Li, Zongyu Lin, Kevin Lin, Guilin Liu, Edith Llontop, Loic Magne, Ajay Mandlekar, Avnish Narayan, Soroush Nasiriany, Scott Reed, You ...

    Pith/arXiv arXiv 2025

  52. [52]

    Scalable diffusion models with transformers.arXiv preprint arXiv: 2212.09748, 2022

    William Peebles and Saining Xie. Scalable diffusion models with transformers.arXiv preprint arXiv: 2212.09748, 2022. 4, 20

    Pith/arXiv arXiv 2022

  53. [53]

    Yarn: Efficient context window extension of large language models

    Bowen Peng, Jeffrey Quesnelle, Honglu Fan, and Enrico Shippole. Yarn: Efficient context window extension of large language models. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024. URLhttps://openreview.net/ forum?id=wHBfxhZu1u. 2

    2024

  54. [54]

    Fast: Efficient action tokenization for vision-language-action models, 2025

    Karl Pertsch, Kyle Stachowicz, Brian Ichter, Danny Driess, Suraj Nair, Quan Vuong, Oier Mees, Chelsea Finn, and Sergey Levine. Fast: Efficient action tokenization for vision-language-action models, 2025. URL https://arxiv.org/abs/2501.09747. 12

    Pith/arXiv arXiv 2025

  55. [55]

    In-hand object rotation via rapid motor adaptation

    Haozhi Qi, Ashish Kumar, Roberto Calandra, Yi Ma, and Jitendra Malik. In-hand object rotation via rapid motor adaptation. InConference on Robot Learning, pages 1722–1732. PMLR, 2023. 11

    2023

  56. [56]

    Scott E. Reed, Konrad Zolna, Emilio Parisotto, Sergio Gómez Colmenarejo, Alexander Novikov, Gabriel Barth-Maron, Mai Gimenez, Yury Sulsky, Jackie Kay, Jost Tobias Springenberg, Tom Eccles, Jake Bruce, Ali Razavi, Ashley Edwards, Nicolas Heess, Yutian Chen, Raia Hadsell, Oriol Vinyals, Mahyar Bordbar, and Nando de Freitas. A generalist agent.Trans. Mach. L...

    2022

  57. [57]

    A reduction of imitation learning and structured prediction to no-regret online learning

    Stephane Ross, Geoffrey Gordon, and Drew Bagnell. A reduction of imitation learning and structured prediction to no-regret online learning. In Geoffrey Gordon, David Dunson, and Miroslav Dudík, editors, Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, volume 15 of Proceedings of Machine Learning Research, p...

    2011

  58. [58]

    Linear transformers are secretly fast weight programmers

    Imanol Schlag, Kazuki Irie, and Jürgen Schmidhuber. Linear transformers are secretly fast weight programmers. InInternational conference on machine learning, pages 9355–9366. PMLR, 2021. 2 17 RoboTTT: Context Scaling for Robot Policies

    2021

  59. [59]

    Eagle: Exploring the design space for multimodal llms with mixture of encoders

    Min Shi, Fuxiao Liu, Shihao Wang, Shijia Liao, Subhashree Radhakrishnan, De-An Huang, Hongxu Yin, Karan Sapra, Yaser Yacoob, Humphrey Shi, Bryan Catanzaro, Andrew Tao, Jan Kautz, Zhiding Yu, and Guilin Liu. Eagle: Exploring the design space for multimodal llms with mixture of encoders. InICLR,

  60. [60]

    Smolvla: A vision-language-action model for affordable and efficient robotics

    Mustafa Shukor, Dana Aubakirova, Francesco Capuano, Pepijn Kooijmans, Steven Palma, Adil Zouitine, Michel Aractingi, Caroline Pascal, Martino Russi, Andres Marafioti, Simon Alibert, Matthieu Cord, Thomas Wolf, and Remi Cadene. Smolvla: A vision-language-action model for affordable and efficient robotics. arXiv preprint arXiv: 2506.01844, 2025. 2, 11

    Pith/arXiv arXiv 2025

  61. [61]

    Memer: Scaling up memory for robot control via experience retrieval.arXiv preprint arXiv: 2510.20328, 2025

    Ajay Sridhar, Jennifer Pan, Satvik Sharma, and Chelsea Finn. Memer: Scaling up memory for robot control via experience retrieval.arXiv preprint arXiv: 2510.20328, 2025. 11, 12

    arXiv 2025

  62. [62]

    Roformer: Enhanced transformer with rotary position embedding.arXiv preprint arXiv: 2104.09864, 2021

    Jianlin Su, Yu Lu, Shengfeng Pan, Ahmed Murtadha, Bo Wen, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.arXiv preprint arXiv: 2104.09864, 2021. 20

    Pith/arXiv arXiv 2021

  63. [63]

    Efros, and Moritz Hardt

    Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei A. Efros, and Moritz Hardt. Test-time training with self-supervision for generalization under distribution shifts. InProceedings of the 37th International Conference on Machine Learning, ICML’20. JMLR.org, 2020. 12, 20

    2020

  64. [64]

    Learning to (learn at test time): Rnns with expressive hidden states.arXiv preprint arXiv: 2407.04620, 2024

    Yu Sun, Xinhao Li, Karan Dalal, Jiarui Xu, Arjun Vikram, Genghan Zhang, Yann Dubois, Xinlei Chen, Xiaolong Wang, Sanmi Koyejo, Tatsunori Hashimoto, and Carlos Guestrin. Learning to (learn at test time): Rnns with expressive hidden states.arXiv preprint arXiv: 2407.04620, 2024. 2, 3

    Pith/arXiv arXiv 2024

  65. [65]

    End-to-end test-time training for long context.arXiv preprint arXiv: 2512.23675, 2025

    Arnuv Tandon, Karan Dalal, Xinhao Li, Daniel Koceja, Marcel Rød, Sam Buchanan, Xiaolong Wang, Jure Leskovec, Sanmi Koyejo, Tatsunori Hashimoto, Carlos Guestrin, Jed McCaleb, Yejin Choi, and Yu Sun. End-to-end test-time training for long context.arXiv preprint arXiv: 2512.23675, 2025. 5, 12

    arXiv 2025

  66. [66]

    Human-timescale adaptation in an open-ended task space.arXiv preprint arXiv:2301.07608, 2023

    Adaptive Agent Team, Jakob Bauer, Kate Baumli, Satinder Baveja, Feryal Behbahani, Avishkar Bhoopc- hand, Nathalie Bradley-Schmieg, Michael Chang, Natalie Clay, Adrian Collister, et al. Human-timescale adaptation in an open-ended task space.arXiv preprint arXiv:2301.07608, 2023. 11

    Pith/arXiv arXiv 2023

  67. [67]

    Sanketi, Quan Vuong, Ted Xiao, Dorsa Sadigh, Chelsea Finn, and Sergey Levine

    Octo Model Team, Dibya Ghosh, Homer Walke, Karl Pertsch, Kevin Black, Oier Mees, Sudeep Dasari, Joey Hejna, Tobias Kreiman, Charles Xu, Jianlan Luo, You Liang Tan, Pannag R. Sanketi, Quan Vuong, Ted Xiao, Dorsa Sadigh, Chelsea Finn, and Sergey Levine. Octo: An open-source generalist robot policy. ROBOTICS, 2024. doi: 10.48550/arXiv.2405.12213. URLhttps://...

    Pith/arXiv arXiv doi:10.48550/arxiv.2405.12213 2024

  68. [68]

    Learning long-context diffusion policies via past-token prediction.arXiv preprint arXiv: 2505.09561, 2025

    Marcel Torne, Andy Tang, Yuejiang Liu, and Chelsea Finn. Learning long-context diffusion policies via past-token prediction.arXiv preprint arXiv: 2505.09561, 2025. 11, 12

    Pith/arXiv arXiv 2025

  69. [69]

    Ren, Haohuan Wang, Jiaming Tang, Kyle Stachowicz, Karan Dhabalia, Michael Equi, Quan Vuong, Jost Tobias Springen- berg, Sergey Levine, Chelsea Finn, and Danny Driess

    Marcel Torne, Karl Pertsch, Homer Walke, Kyle Vedder, Suraj Nair, Brian Ichter, Allen Z. Ren, Haohuan Wang, Jiaming Tang, Kyle Stachowicz, Karan Dhabalia, Michael Equi, Quan Vuong, Jost Tobias Springen- berg, Sergey Levine, Chelsea Finn, and Danny Driess. Mem: Multi-scale embodied memory for vision language action models.arXiv preprint arXiv: 2603.03596, ...

    arXiv 2026

  70. [70]

    Fighting copycat agents in behavioral cloning from observation histories.arXiv preprint arXiv: 2010.14876, 2020

    Chuan Wen, Jierui Lin, Trevor Darrell, Dinesh Jayaraman, and Yang Gao. Fighting copycat agents in behavioral cloning from observation histories.arXiv preprint arXiv: 2010.14876, 2020. 11

    Pith/arXiv arXiv 2010

  71. [71]

    Efficient streaming language models with attention sinks

    Guangxuan Xiao, Yuandong Tian, Beidi Chen, Song Han, and Mike Lewis. Efficient streaming language models with attention sinks. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024. URLhttps://openreview.net/forum? id=NG7sS51zVF. 2

    2024

  72. [72]

    Magma: A foundation model for multimodal ai agents, 2025

    Jianwei Yang, Reuben Tan, Qianhui Wu, Ruijie Zheng, Baolin Peng, Yongyuan Liang, Yu Gu, Mu Cai, Seonghyeon Ye, Joel Jang, Yuquan Deng, Lars Liden, and Jianfeng Gao. Magma: A foundation model for multimodal ai agents, 2025. URLhttps://arxiv.org/abs/2502.13130. 12 18 RoboTTT: Context Scaling for Robot Policies

    Pith/arXiv arXiv 2025

  73. [73]

    Fla: A triton-based library for hardware-efficient implementations of linear attention mechanism, January 2024

    Songlin Yang and Yu Zhang. Fla: A triton-based library for hardware-efficient implementations of linear attention mechanism, January 2024. URLhttps://github.com/fla-org/flash-linear-attention. 22

    2024

  74. [74]

    Gated delta networks: Improving mamba2 with delta rule.International Conference on Learning Representations, 2024

    Songlin Yang, Jan Kautz, and Ali Hatamizadeh. Gated delta networks: Improving mamba2 with delta rule.International Conference on Learning Representations, 2024. doi: 10.48550/arXiv.2412.06464. 7

    Pith/arXiv arXiv doi:10.48550/arxiv.2412.06464 2024

  75. [75]

    World action models are zero-shot policies.arXiv preprint arXiv:2602.15922, 2026

    Seonghyeon Ye, Yunhao Ge, Kaiyuan Zheng, Shenyuan Gao, Sihyun Yu, George Kurian, Suneel Indupuru, You Liang Tan, Chuning Zhu, Jiannan Xiang, et al. World action models are zero-shot policies.arXiv preprint arXiv:2602.15922, 2026. 2, 11

    Pith/arXiv arXiv 2026

  76. [76]

    Fast-wam: Do world action models need test-time future imagination?arXiv preprint arXiv: 2603.16666, 2026

    Tianyuan Yuan, Zibin Dong, Yicheng Liu, and Hang Zhao. Fast-wam: Do world action models need test-time future imagination?arXiv preprint arXiv: 2603.16666, 2026. 2, 11

    Pith/arXiv arXiv 2026

  77. [77]

    Learning to discover at test time, 2026

    Mert Yuksekgonul, Daniel Koceja, Xinhao Li, Federico Bianchi, Jed McCaleb, Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou, Carlos Guestrin, and Yu Sun. Learning to discover at test time, 2026. URL https://arxiv.org/abs/2601.16175. 12

    Pith/arXiv arXiv 2026

  78. [78]

    Freeman, and Hao Tan

    Tianyuan Zhang, Sai Bi, Yicong Hong, Kai Zhang, Fujun Luan, Songlin Yang, Kalyan Sunkavalli, William T. Freeman, and Hao Tan. Test-time training done right.arXiv preprint arXiv: 2505.23884, 2025. 2, 3, 11, 12, 20

    Pith/arXiv arXiv 2025

  79. [79]

    Fast-weight product key memory.arXiv preprint arXiv: 2601.00671, 2026

    Tianyu Zhao and Llion Jones. Fast-weight product key memory.arXiv preprint arXiv: 2601.00671, 2026. 12

    arXiv 2026

  80. [80]

    TraceVLA: Visual trace prompting enhances spatial-temporal awareness for generalist robotic policies

    Ruijie Zheng, Yongyuan Liang, Shuaiyi Huang, Jianfeng Gao, Hal Daumé III, Andrey Kolobov, Furong Huang, and Jianwei Yang. TraceVLA: Visual trace prompting enhances spatial-temporal awareness for generalist robotic policies. InThe Thirteenth International Conference on Learning Representations, 2025. 11, 12

    2025

Showing first 80 references.