Pith. sign in

REVIEW 5 major objections 7 minor 49 references

Long-Horizon Embodied Decision-Making via Multimodal Memory Compression

T0 review · 5 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper argues that long-horizon embodied decision-making fails from too much unfiltered multimodal history, not too little context, and that preference-conditioned compression of memory into fixed tokens improves accuracy by 7.18% while

desk verdict A genuinely useful benchmark idea and a sensible memory compressor, but the main accuracy claim is undermined by an unreported evaluation inconsistency and a tiny test set. read the letter →

arxiv 2608.01456 v1 pith:7BY3WBCH submitted 2026-08-02 cs.CV cs.CL

Bingxuan Li , Rui Yang , Cheng Qian , Jiateng Liu , Jeonghwan Kim , Zhenhailong Wang , Manling Li , Tong Zhang
show 1 more author
Heng Ji
This is my paper
classification cs.CVcs.CL
keywords embodieddecision-makingmultimodalmemorycompressionvision-languagemodelspreferencealignmentlong-horizonreasoningbenchmarkevaluationuserpreferences
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper is trying to establish that long-horizon, preference-driven embodied decision-making is bottlenecked not by how much history an agent can hold but by how it compresses that history. On a new benchmark, DunphyBench, agents tour multiple homes, accumulate multimodal observations, and pick the home best matching a user's explicit or implicit preferences; the best agent reaches 58.3% accuracy versus 83.3% for humans. Diagnosing failures, the authors find memory management is a key bottleneck: accuracy rises with history up to about 10 turns and then declines as unfiltered context adds noise. Their proposed MeMento compressor conditions a fixed set of memory tokens on user preferences and compresses per-environment step history, improving VLM-driven agent accuracy by 7.18% absolute while reducing memory usage by 85.38% compared with the strongest baseline.

What carries the argument

The central object is a preference-conditioned Perceiver-style compressor: a fixed set of learnable memory queries $Q_0$, shifted by a projection of a pooled preference embedding $\bar{z}_r$ to form $Q_r$, cross-attend over step embeddings of each environment separately, then self-attend to produce environment-level memory tokens, which are concatenated and projected into a frozen VLM's embedding space. The mechanism's job is to make the retained history depend on what the user cares about, so irrelevant or redundant multimodal observations are discarded rather than accumulated.

What would settle it

Run a larger evaluation (e.g., several hundred instances rather than roughly 18 per track) and check whether MeMento's 7.18% absolute accuracy gain over the strongest baseline shrinks to statistical noise. Additionally, replace MeMento with an ordinary retriever that selects preference-relevant steps and uses them as context; if retrieval matches MeMento's accuracy, the learned compression mechanism is not the causal source of the gain.

Watch

Extended reading notes

Core claim

MeMento is a preference-conditioned multimodal memory compressor. Each environment's step history is encoded into multimodal embeddings, augmented with step and environment identifiers, then compressed by a fixed set of learned queries that have been shifted by a pooled preference embedding. Cross-attention gathers decision-relevant evidence per environment, self-attention fuses across steps, and an adapter injects the compressed tokens into a frozen VLM as a soft prompt. The compressor is trained in two stages: first to predict step-level preference relevance and align cross-attention with relevance labels, then to answer preference-related memory-probe questions so the compressed tokens re

Load-bearing premise

The load-bearing assumption is that every implicit natural-language preference can be unambiguously mapped to one deterministic predicate over the structured feature vector; if that mapping is artificial, the measured gains may reflect simplified keyword-to-predicate matching rather than human-aligned preference reasoning.

Editorial extensions

If this is right

  • Agents should manage memory by preference relevance rather than by recency or raw context length.
  • Because accuracy degrades beyond roughly 10 turns of unfiltered history, simply extending the context window will not solve long-horizon embodied decision-making.
  • Compressed, environment-separated memory can match or beat much larger multimodal memory at a fraction of the token cost.
  • The same preference-conditioned compression transfers to web-agent decision tasks, where MeMento reaches 83% success with 512 tokens.

Reading between the lines

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

  • We infer that if the non-monotonic memory curve generalizes, the main practical task for embodied assistants shifts from storing more to deciding what to discard; retrieval alone may be insufficient because it still presents potentially noisy context to the model.
  • We also infer that the reported 7.18% absolute gain rests on a small test set (about 18 instances per track), so the precise magnitude is fragile; a larger sample is needed before treating the improvement as stable.
  • We further infer that because the benchmark grounds implicit preferences in deterministic predicates, the implicit track may understate the ambiguity of real human preference; richer commonsense inference would be a harder test for MeMento.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 7 minor

Summary. The paper introduces DunphyBench, a benchmark for long-horizon embodied decision-making in which an agent explores multiple housing environments and selects the one best matching a set of explicit or implicit user preferences. The authors evaluate several VLM-driven agents, report a substantial gap to human performance, and identify memory management as a key bottleneck via a diagnosis showing non-monotonic accuracy as context history grows. They then propose MeMento, a preference-conditioned multimodal memory compressor based on a Perceiver-style module with learned memory queries, trained in two stages (relevance prediction and alignment with a frozen VLM). The central claimed results are that MeMento improves decision accuracy by 7.18% absolute (15.74% relative) over the strongest baseline on DunphyBench while reducing memory usage by 85.38%. The paper also reports ablations showing all components contribute, a memory-probing QA evaluation, and a preliminary transfer result to web-agent tasks.

Significance. If the central claims hold, the paper makes two useful contributions: a reproducible, automatically verifiable benchmark for preference-driven long-horizon embodied decision-making, and a memory-compression method that improves both accuracy and token efficiency for VLM agents. The benchmark's use of deterministic predicates for automatic verification is a genuine strength, as is the explicit separation of explicit and implicit preference tracks. The diagnostic finding that memory scaling is non-monotonic is also valuable. However, the headline quantitative claims rest on an evaluation figure whose numbers are inconsistent with the main benchmark table, on a very small test set, and on unspecified train/test separation for MeMento's training labels. These issues are load-bearing for the paper's main claims and must be resolved before the results can be assessed.

major comments (5)
  1. [§5.6.1, Figure 5 vs. Table 2] The central claim that MeMento improves accuracy by 7.18% absolute over the strongest baseline is supported only by Figure 5, which reports Qwen3VL-8B baselines of 26% (Memoryless), 39% (Text Mem), and 30% (Multimodal Mem). These values do not match the corresponding Qwen3VL-8B rows in Table 2, which report 8.3%, 13.9%, and 16.7% for the same configurations. Section 5.6.1 states the experiments use 'same protocol as Section 4,' but no split, difficulty tier, or filtering rule is given for Figure 5. If Figure 5 was run on a different, easier, or overlapping subset, the claimed improvement over the 'strongest baseline' is not established on DunphyBench as defined in Section 4.
  2. [§4.1, Table 2] The accuracy percentages in Table 2 are consistent with a total test set of approximately 36 instances (18 per track): for example, 8.3% = 3/36 and 58.3% = 21/36. A 7.18% absolute improvement therefore corresponds to roughly 2.6 instances. The paper reports no confidence intervals, significance tests, or per-instance predictions. The headline gain over the strongest baseline may be sampling noise; at minimum, the statistical uncertainty should be quantified before claiming a robust improvement.
  3. [§5.4, §5.6.1] MeMento is trained with step-level relevance labels, and those labels are constructed from the same preference-predicate definitions used to determine correctness on DunphyBench. The paper does not specify whether MeMento's training data come from the same instances that are subsequently evaluated, or from a separate training split. Since the main experiment is on DunphyBench and Section 5.6.1 only says 'same protocol as Section 4,' the possibility of training on the evaluation instances is not ruled out. The paper must specify the split, the number of training trajectories, and how the relevance-label supervision avoids overlap with the test set.
  4. [§3.2, Step 5] The data-generation pipeline requires that each implicit statement 'uniquely imply a single predicate.' This design choice makes automatic verification possible, but it also means the implicit track reduces to a one-to-one natural-language-to-predicate mapping. Under this mapping, 'implicit preference reasoning' may be keyword-to-predicate matching rather than reasoning about underspecified intent. The paper's claims about implicit-preference capability gaps (e.g., Insight 2) should be tempered or validated with a human study showing that the unique mapping is natural and not an artifact of the generation pipeline.
  5. [§4.2.2, Figure 3(right)] The diagnosis that accuracy improves with history up to 10 turns and then declines is reported only as a qualitative statement ('improves accuracy by 17%') with a figure. No error bars, no per-turn counts, and no significance testing are given. Since this diagnosis motivates the entire MeMento method, it should be presented with the same rigor as the main evaluation, including the number of instances per turn-budget condition and confidence intervals.
minor comments (7)
  1. [Throughout] The manuscript contains many unresolved 'Appendix ??' references (e.g., Sections 3.1, 3.2, 4.1, 5.6.1, 6.1, 6.2, 6.3). The appendix is absent in the submitted version, making several setup details unverifiable.
  2. [§6.2] The text refers to 'Table 5' for the memory-probing results, but the results appear only as a table embedded in Figure 5 (Right). This should be a numbered table or the reference should be corrected.
  3. [§5.1] Typo: 'We then encode encode each explored step'.
  4. [§5.4] Typo: 'an relevance label' should be 'a relevance label'; also 'L stage1 = L rel + λ L attn' is missing the numeral and spacing in the displayed equation.
  5. [§6.3] Typo: 'bassline' should be 'baseline'.
  6. [Table 2] The symbols for memory and tool configurations (e.g., /times, /f⌢nt, ⊷, ♂pen) are not legible in a standard PDF and are not defined in the table caption. Standard textual abbreviations or a legend would greatly improve readability.
  7. [§1, first paragraph] 'the option that best matches their needs' has a pronoun mismatch ('their' referring to singular 'you'); minor wording issue.

Circularity Check

1 steps flagged · score 2.0 of 10

One supporting analysis (memory probing) is circular because MeMento is trained on the same preference-QA objective it is evaluated on; the main DunphyBench accuracy claim is independent.

  1. fitted input called prediction [Section 5.4 Stage 2 (training) and Section 6.2 (memory probing evaluation)]
    "In this stage, we train the compressor and projection adapter to make the compressed memory readable by the frozen VLM. We construct the training dataset with preference-related QA questions, such as whether an object was observed in a room, whether a relation was absent. ... We design a memory probing task... We therefore ask the agent to answer follow-up preference-related questions about previously explored environments using only its stored memory. ... MeMento achieves the best non-oracle macro-F1 score (0.719), outperforming other baselines."

    MeMento is explicitly trained in Stage 2 to answer preference-related QA with ground-truth answers derived from the same environment representations used for DunphyBench verification. The memory probing evaluation in §6.2 asks the same class of preference-related questions (e.g., 'Was there a window near the sink?') and reports MeMento's F1 as evidence that compression preserves decision-relevant evidence. Because the compressor's parameters are optimized for exactly this task, while the baselines (text recent, summary, multimodal recent) receive no such training, the measured gap is a direct consequence of fitting to the probing objective. The probing result is therefore a fitted-input prediction, not an independent confirmation of a general memory-preservation ability.

full rationale

The paper's central claim—MeMento improves final decision accuracy on DunphyBench—is not circular in the strict sense. The decision accuracy is measured on held-out instances by comparing the chosen environment against the golden environment, and the compressor's training objectives (step relevance and preference-QA) are not identical to the final decision target. The web-agent transfer experiment (§6.3) provides external grounding beyond DunphyBench. The self-citations present in the references are not load-bearing for the main derivation. However, the memory probing analysis in §6.2 is circular: MeMento is trained to answer preference-related QA in Stage 2 and then evaluated on the same type of QA, while baselines are not trained for that objective, making the favorable F1 comparison partially forced by construction. This issue is confined to a supporting analysis and does not invalidate the main decision-accuracy result, so the circularity score is low. Note that the evaluation inconsistency between Figure 5 and Table 2 is a correctness/reproducibility concern, not a circularity concern per the review criteria.

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

The central result depends on several unvalidated domain assumptions about predicate grounding, implicit-preference uniqueness, and the transferability of soft-prompt memory. No new physical entities are introduced. The learned parameters of MeMento and the hand-chosen benchmark thresholds are the main fitted quantities, and their values are largely unspecified.

free parameters (4)
  • Number of learned memory queries/tokens Q0 = unspecified
    The fixed set of learnable queries used for compression; the count and dimension are not reported in the main text and directly affect accuracy and memory cost.
  • Stage-1 loss weight lambda = unspecified
    Scales the relevance cross-entropy loss and attention loss in Section 5.4; the value is not stated.
  • Distractor satisfaction thresholds = easy <0.3N, medium 0.3-0.6N, hard 0.6-0.9N, >=0.9N excluded
    Hand-chosen difficulty tiers in Section 3.2 Step 4 that define the distractor pool and therefore the measured accuracy.
  • Default benchmark configuration N and M = N=7, M=19
    The default number of preference requirements and distractors per tier, set in Section 4.1 and used for the main evaluation.
assumptions (6)
  • domain assumption Structured feature vectors from ProcTHOR and deterministic predicates faithfully represent residential environments and human preferences.
    Section 2 states that all preferences must be grounded in a representation that supports automatic verification; this assumes no semantic gap between natural-language preferences and the predefined predicates.
  • domain assumption LLM-generated implicit preference statements are valid and each uniquely implies exactly one predicate.
    Section 3.2 Step 3 requires that 'each statement must uniquely imply a single predicate'; if this is artificial, the implicit track does not measure genuine preference inference.
  • domain assumption Human annotator performance under the same protocol is a reliable 83.3% accuracy upper bound.
    Section 4.1 says humans are evaluated with the same protocol, but the number of annotators, instructions, and quality controls are not described.
  • domain assumption A frozen VLM can be steered toward better decisions by injecting trainable soft-prompt memory tokens.
    Section 5.3 assumes the projection adapter can map compressor outputs into the VLM embedding space without changing the backbone.
  • domain assumption Stage-2 QA alignment training transfers to next-action decision accuracy.
    MeMento is trained with preference-related question answering in Stage 2, not with final decision labels; the paper assumes this alignment improves downstream choices.
  • domain assumption The 100 Embodied Web Agent tasks are a valid domain-transfer test for the compressor.
    Section 6.3 reports 83% success on 100 web tasks, but the task composition, scoring, and relation to housing preferences are not detailed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Long-Horizon Embodied Decision-Making via Multimodal Memory Compression." pith.science (2026). https://pith.science/paper/7BY3WBCH

@misc{pith2026260801456,
  author       = {Pith},
  title        = {Pith review of: Long-Horizon Embodied Decision-Making via Multimodal Memory Compression},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7BY3WBCH}},
  note         = {Machine review of arXiv:2608.01456}
}
read the original abstract

Agents are increasingly expected to act not only as task executors, but also as decision-makers on behalf of human users. This shift requires agents to accumulate evidence over long horizons, interpret implicit user preferences, and compare multiple candidates under partial observations. In this work, we propose DunphyBench, a new benchmark for evaluating agents on long-horizon human-centered embodied decision-making, where the agent must navigate through multiple embodied housing environments and make decisions that align with multi-dimensional human preferences. Unlike standard embodied reasoning tasks that often focus on procedural planning or immediate goal completion, our setting requires agents to integrate multimodal, multi-source input into coherent knowledge that supports complex reasoning across long horizon. The evaluation results reveal that there is a substantial gap between current agents and human performance. Furthermore, our diagnosis of state-of-the-art VLM-driven agents reveals that memory management is one of the bottlenecks, where raw multimodal history introduces noise that hinders decision quality. Motivated by this finding, we design MeMento, a preference-conditioned multimodal memory compressor that selectively compresses decision-relevant information from long-horizon history based on user preferences with a fixed set of memory tokens. Experiments show that MeMento helps VLM-driven agents improve accuracy by 7.18%, while reducing memory usage by 85.38% compared to the strongest baseline.

Figures

Figures reproduced from arXiv: 2608.01456 by the authors.

Figure 1
Figure 1. Task illustration of DunphyBench. To systematically study this problem, we in￾troduce DunphyBench, a new benchmark to challenge agents with the task of long-horizon human-centered embodied decision-making, where the agent must navigate through embod￾ied household environments and make decisions that align with compositional human prefer￾ences. The benchmark includes both explicit preferences, expressed as direct req… view at source ↗
Figure 2
Figure 2. Performance of Qwen3VL-8B-Instruct agent under varying task settings. Left: Accuracy across difficulty levels. Right: Impact of increasing number of requirements(N) and distractors(M). M = 19 reduces accuracy from 42.6% to 15.3%. These results suggest that agents are sensitive to compositional complexity and cluttered decision spaces. Insight 2. Agents struggle to interpret implicit preferences. As illustrated in [… view at source ↗
Figure 3
Figure 3. Performance analysis on efficiency and agent configuration. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Overview of MeMento: Preference-conditioned multimodal memory compression. We then encode encode each explored step in multimodal memory into a multimodal step representa￾tion. Each step contains the ego-view image, map view, observation text, and action text: xt = {eg…
Figure 5
Figure 5. Figure 5: MeMento experiment results. Left: Main Experiment. Right: Memory Probing QA using only textual memory, excluding visual information; (3) MeMentoU : removes the hierarchical two-stage design and instead compresses the entire memory in a single stage. Results. All compon…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

49 extracted references · 26 canonical work pages

  1. [1]

    Deitke, Matt and VanderBilt, Eli and Herrasti, Alvaro and Weihs, Luca and Salvador, Jordi and Ehsani, Kiana and Han, Winson and Kolve, Eric and Farhadi, Ali and Kembhavi, Aniruddha and Mottaghi, Roozbeh , booktitle=

    work page

  2. [2]

    2021 , url =

    Mohit Shridhar and Xingdi Yuan and Marc-Alexandre C\^ot\'e and Yonatan Bisk and Adam Trischler and Matthew Hausknecht , booktitle =. 2021 , url =

    2021

  3. [3]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Alfred: A benchmark for interpreting grounded instructions for everyday tasks , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  4. [5]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Embodied question answering , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  5. [10]

    Advances in Neural Information Processing Systems , volume=

    Embodied agent interface: Benchmarking llms for embodied decision making , author=. Advances in Neural Information Processing Systems , volume=

  6. [11]

    Conference on Robot Learning , pages=

    Behavior-1k: A benchmark for embodied ai with 1,000 everyday activities and realistic simulation , author=. Conference on Robot Learning , pages=. 2023 , organization=

    2023

  7. [13]

    Journal of health psychology , volume=

    Decision fatigue: A conceptual analysis , author=. Journal of health psychology , volume=. 2020 , publisher=

    work page 2020

  8. [14]

    2024 , institution=

    GPT-4o System Card , author=. 2024 , institution=

    work page 2024

Show all 49 references
  1. [15]

    2026 , institution=

    GPT-5.4 Thinking System Card , author=. 2026 , institution=

    2026

  2. [16]

    arXiv preprint arXiv:2507.06261 , year=

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities , author=. arXiv preprint arXiv:2507.06261 , year=

    arXiv

  3. [17]

    arXiv preprint arXiv:2511.21631 , year=

    Qwen3-VL Technical Report , author=. arXiv preprint arXiv:2511.21631 , year=

    arXiv

  4. [18]

    2025 , eprint=

    MolmoAct: Action Reasoning Models that can Reason in Space , author=. 2025 , eprint=

    2025

  5. [20]

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

    Metal: A multi-agent framework for chart generation with test-time scaling , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  6. [21]

    2026 , eprint=

    OSExpert: Computer-Use Agents Learning Professional Skills via Exploration , author=. 2026 , eprint=

    2026

  7. [23]

    Zhang, Shiduo and Xu, Zhe and Liu, Peiju and Yu, Xiaopeng and Li, Yuan and Gao, Qinghui and Fei, Zhaoye and Yin, Zhangyue and Wu, Zuxuan and Jiang, Yu-Gang and Qiu, Xipeng , booktitle =

  8. [24]

    arXiv preprint arXiv:2411.00081 , year=

    Partnr: A benchmark for planning and reasoning in embodied multi-agent tasks , author=. arXiv preprint arXiv:2411.00081 , year=

    arXiv

  9. [25]

    V esta B ench: An Embodied Benchmark for Safe Long-Horizon Planning Under Multi-Constraint and Adversarial Settings

    Sadhu, Tanmana and Chen, Yanan and Pesaranghader, Ali. V esta B ench: An Embodied Benchmark for Safe Long-Horizon Planning Under Multi-Constraint and Adversarial Settings. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track. 2...

    2025 doi

  10. [26]

    OpenEQA: Embodied Question Answering in the Era of Foundation Models , booktitle =

    Majumdar, Arjun and Ajay, Anurag and Zhang, Xiaohan and Putta, Pranav and Yenamandra, Sriram and Henaff, Mikael and Silwal, Sneha and Mcvay, Paul and Maksymets, Oleksandr and Arnaud, Sergio and Yadav, Karmesh and Li, Qiyang and Newman, Ben and Sharma, Mohit and Berges, Vincent...

  11. [27]

    2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Is the house ready for sleeptime? Generating and evaluating situational queries for embodied question answering , author=. 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2025 , organization=

    2025

  12. [28]

    arXiv preprint arXiv:2403.15941 , year=

    Explore until confident: Efficient exploration for embodied question answering , author=. arXiv preprint arXiv:2403.15941 , year=

    arXiv

  13. [29]

    2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Efficienteqa: An efficient approach to open-vocabulary embodied question answering , author=. 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2025 , organization=

    2025

  14. [30]

    2025 , doi =

    Zhao, Yong and Xu, Kai and Zhu, Zhengqiu and Hu, Yue and Zheng, Zhiheng and Chen, Yingfeng and Ji, Yatai and Gao, Chen and Li, Yong and Huang, Jincai , booktitle =. 2025 , doi =

    2025

  15. [31]

    2025 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Remembr: Building and reasoning over long-horizon spatio-temporal memory for robot navigation , author=. 2025 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2025 , organization=

    2025

  16. [32]

    2025 , publisher =

    Saxena, Saumya and Buchanan, Blake and Paxton, Chris and Liu, Peiqi and Chen, Bingqing and Vaskevicius, Narunas and Palmieri, Luigi and Francis, Jonathan and Kroemer, Oliver , booktitle =. 2025 , publisher =

    2025

  17. [33]

    Proceedings of The 9th Conference on Robot Learning , series =

    Enter the Mind Palace: Reasoning and Planning for Long-term Active Embodied Question Answering , author =. Proceedings of The 9th Conference on Robot Learning , series =. 2025 , publisher =

    2025

  18. [34]

    arXiv preprint arXiv:2602.15513 , year=

    HIMM: Human-Inspired Long-Term Memory Modeling for Embodied Exploration and Question Answering , author=. arXiv preprint arXiv:2602.15513 , year=

    arXiv

  19. [35]

    Advances in neural information processing systems , volume=

    Efficient large multi-modal models via visual context compression , author=. Advances in neural information processing systems , volume=

  20. [36]

    Findings of the Association for Computational Linguistics: NAACL 2025 , pages=

    Lvpruning: An effective yet simple language-guided vision token pruning approach for multi-modal large language models , author=. Findings of the Association for Computational Linguistics: NAACL 2025 , pages=

    2025

  21. [37]

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

    Atp-llava: Adaptive token pruning for large vision language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  22. [38]

    arXiv preprint arXiv:2501.00574 , year=

    Videochat-flash: Hierarchical compression for long-context video modeling , author=. arXiv preprint arXiv:2501.00574 , year=

    arXiv

  23. [39]

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

    Zero-shot 3d question answering via voxel-based dynamic token compression , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  24. [40]

    arXiv preprint arXiv:2603.15167 , year=

    Question-guided Visual Compression with Memory Feedback for Long-Term Video Understanding , author=. arXiv preprint arXiv:2603.15167 , year=

  25. [41]

    arXiv preprint arXiv:2603.02096 , year=

    FluxMem: Adaptive Hierarchical Memory for Streaming Video Understanding , author=. arXiv preprint arXiv:2603.02096 , year=

  26. [42]

    arXiv preprint arXiv:2604.09000 , year=

    Streammeco: Long-term agent memory compression for efficient streaming video understanding , author=. arXiv preprint arXiv:2604.09000 , year=

    arXiv

  27. [43]

    Babyai: A platform to study the sample efficiency of grounded language learning

    Maxime Chevalier-Boisvert, Dzmitry Bahdanau, Salem Lahlou, Lucas Willems, Chitwan Saharia, Thien Huu Nguyen, and Yoshua Bengio. Babyai: A platform to study the sample efficiency of grounded language learning. arXiv preprint arXiv:1810.08272, 2018

    2018 arXiv

  28. [44]

    ProcTHOR : Large-scale embodied ai using procedural generation

    Matt Deitke, Eli VanderBilt, Alvaro Herrasti, Luca Weihs, Jordi Salvador, Kiana Ehsani, Winson Han, Eric Kolve, Ali Farhadi, Aniruddha Kembhavi, and Roozbeh Mottaghi. ProcTHOR : Large-scale embodied ai using procedural generation. In Advances in Neural Information Processing S...

    2022

  29. [45]

    Embodied web agents: Bridging physical-digital realms for integrated agent intelligence

    Yining Hong, Rui Sun, Bingxuan Li, Xingcheng Yao, Maxine Wu, Alexander Chien, Da Yin, Ying Nian Wu, Zhecan James Wang, and Kai-Wei Chang. Embodied web agents: Bridging physical-digital realms for integrated agent intelligence. arXiv preprint arXiv:2506.15677, 2025

    2025 arXiv

  30. [46]

    3dllm-mem: Long-term spatial-temporal memory for embodied 3d large language model

    Wenbo Hu, Yining Hong, Yanjun Wang, Leison Gao, Zibu Wei, Xingcheng Yao, Nanyun Peng, Yonatan Bitton, Idan Szpektor, and Kai-Wei Chang. 3dllm-mem: Long-term spatial-temporal memory for embodied 3d large language model. arXiv preprint arXiv:2505.22657, 2025

    2025

  31. [47]

    Echofoley: Event-centric hierarchical control for video grounded creative sound generation

    Bingxuan Li, Yiming Cui, Yicheng He, Yiwei Wang, Shu Zhang, Longyin Wen, and Yulei Niu. Echofoley: Event-centric hierarchical control for video grounded creative sound generation. arXiv preprint arXiv:2512.24731, 2025 a

    2025 arXiv

  32. [48]

    Metal: A multi-agent framework for chart generation with test-time scaling

    Bingxuan Li, Yiwei Wang, Jiuxiang Gu, Kai-Wei Chang, and Nanyun Peng. Metal: A multi-agent framework for chart generation with test-time scaling. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 30054--30...

    2025

  33. [49]

    Pearl: Self-evolving assistant for time management with reinforcement learning

    Bingxuan Li, Jeonghwan Kim, Cheng Qian, Xiusi Chen, Eitan Anzenberg, Niran Kundapur, and Heng Ji. Pearl: Self-evolving assistant for time management with reinforcement learning. arXiv preprint arXiv:2601.11957, 2026

    2026 arXiv

  34. [50]

    Embodied agent interface: Benchmarking llms for embodied decision making

    Manling Li, Shiyu Zhao, Qineng Wang, Kangrui Wang, Yu Zhou, Sanjana Srivastava, Cem Gokmen, Tony Lee, Li E Li, Ruohan Zhang, et al. Embodied agent interface: Benchmarking llms for embodied decision making. Advances in Neural Information Processing Systems, 37: 0 100428--100534, 2024

    2024

  35. [51]

    Osexpert: Computer-use agents learning professional skills via exploration, 2026

    Jiateng Liu, Zhenhailong Wang, Rushi Wang, Bingxuan Li, Jeonghwan Kim, Aditi Tiwari, Pengfei Yu, Denghui Zhang, and Heng Ji. Osexpert: Computer-use agents learning professional skills via exploration, 2026. URL https://arxiv.org/abs/2603.07978

    2026

  36. [52]

    Decision fatigue: A conceptual analysis

    Grant A Pignatiello, Richard J Martin, and Ronald L Hickman Jr. Decision fatigue: A conceptual analysis. Journal of health psychology, 25 0 (1): 0 123--135, 2020

    2020

  37. [53]

    Userbench: An interactive gym environment for user-centric agents

    Cheng Qian, Zuxin Liu, Akshara Prabhakar, Zhiwei Liu, Jianguo Zhang, Haolin Chen, Heng Ji, Weiran Yao, Shelby Heinecke, Silvio Savarese, et al. Userbench: An interactive gym environment for user-centric agents. arXiv preprint arXiv:2507.22034, 2025

    2025 arXiv

  38. [54]

    Alfred: A benchmark for interpreting grounded instructions for everyday tasks

    Mohit Shridhar, Jesse Thomason, Daniel Gordon, Yonatan Bisk, Winson Han, Roozbeh Mottaghi, Luke Zettlemoyer, and Dieter Fox. Alfred: A benchmark for interpreting grounded instructions for everyday tasks. In Proceedings of the IEEE/CVF conference on computer vision and pattern ...

    2020

  39. [55]

    ALFWorld: Aligning Text and Embodied Environments for Interactive Learning

    Mohit Shridhar, Xingdi Yuan, Marc-Alexandre C\^ot\'e, Yonatan Bisk, Adam Trischler, and Matthew Hausknecht. ALFWorld: Aligning Text and Embodied Environments for Interactive Learning . In Proceedings of the International Conference on Learning Representations (ICLR), 2021. URL...

    2021 arXiv

  40. [56]

    Embodiedbench: Comprehensive benchmarking multi-modal large language models for vision-driven embodied agents

    Rui Yang, Hanyang Chen, Junyu Zhang, Mark Zhao, Cheng Qian, Kangrui Wang, Qineng Wang, Teja Venkat Koripella, Marziyeh Movahedi, Manling Li, et al. Embodiedbench: Comprehensive benchmarking multi-modal large language models for vision-driven embodied agents. arXiv preprint arX...

    2025 arXiv

  41. [57]

    Webarena: A realistic web environment for building autonomous agents

    Shuyan Zhou, Frank F Xu, Hao Zhu, Xuhui Zhou, Robert Lo, Abishek Sridhar, Xianyi Cheng, Tianyue Ou, Yonatan Bisk, Daniel Fried, et al. Webarena: A realistic web environment for building autonomous agents. arXiv preprint arXiv:2307.13854, 2023

    2023 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.