Pith. sign in

REVIEW 3 major objections 7 minor 52 references

Robo-DM: Data Management For Large Robot Datasets

T0 review · 3 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A robot-data format that compresses demonstrations up to 73x with no loss in a physical policy test.

desk verdict Solid data-format engineering with real compression and loading numbers; the blanket 'no accuracy loss' claim outruns the evidence. read the letter →

arxiv 2505.15558 v1 pith:A5D7DI5Z submitted 2025-05-21 cs.RO cs.AIcs.DBcs.LG

Kaiyuan Chen , Letian Fu , David Huang , Yanxiang Zhang , Lawrence Yunliang Chen , Huang Huang , Kush Hari , Ashwin Balakrishna
show 4 more authors
Ted Xiao Pannag R Sanketi John Kubiatowicz Ken Goldberg
This is my paper · ORCID
classification cs.ROcs.AIcs.DBcs.LG
keywords robotdatamanagementEBMLlossyvideocompressionloadingimitationlearningOpen-X-Embodimenttransformerpolicies
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

Robo-DM proposes a new way to store, share, and load large collections of robot demonstration data. The paper argues that a self-contained container format based on EBML, combined with lossy video compression for camera streams, can shrink dataset sizes by 18x to 73x compared with the RLDS format used by Open-X-Embodiment while keeping training data loading fast. It also claims that models trained on this compressed data do not lose downstream task accuracy: a pick-and-place policy trained on 75.3x-compressed data succeeded in 15 of 15 physical trials, and an Octo fine-tune showed only a 2.6% increase in validation MSE. If these results hold, the format could lower the cost and bandwidth barriers that currently limit sharing and using large robot datasets.

What carries the argument

The load-bearing object is the Robo-DM file format: an extension of Extensible Binary Meta Language (EBML), a binary markup language similar to XML but compact, used by the MKV video container. Robo-DM wraps each vision, language, and action data packet with a timestamp and multiplexes the streams into one self-contained file, so an episode can be read without assumptions about alignment. On top of this container, Robo-DM applies off-the-shelf video codecs (H.264, H.265, AV1) for lossy compression, and uses memory-mapped files with a load-balancing heuristic to cache decoded frames in RAM and avoid redundant video decoding during training. This arrangement is what lets the framework claim both large space savings and fast random access to episodes.

What would settle it

Train the same In-Context Robot Transformer on the same 335 demonstrations stored losslessly (for example in original HDF5 or Robo-DM lossless), run the same 15 randomized pick-and-place trials, and check whether the lossless-trained model achieves a higher success rate or more consistent grasps than the 15/15 lossy-trained model; if it does, the claim that 75x lossy compression does not reduce downstream accuracy is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that robot trajectory data can be stored as a single self-contained EBML container that multiplexes video, language, and action streams with relative timestamps, and that this design makes robot datasets dramatically smaller and faster to load than existing frameworks. In experiments on four Open-X-Embodiment datasets, Robo-DM reduced per-episode size by 18x to 73x relative to RLDS, and loaded episodes up to 33x faster than LeRobot (with a separate benchmark reporting up to 50x faster sequential decoding). The authors verify that the lossy AV1 compression at CRF 30 does not ruin downstream learning: fine-tuning Octo on a 4.39x-compressed RT-1 subset increased validation image-conditioned MSE from 1.86 to 1.91, and training In-Context Robot Transformer on 335 demonstrations compressed 75.3x yielded 15/15 success on a physical pick-and-place task.

Load-bearing premise

The load-bearing premise is that AV1 lossy video at CRF 30 preserves the visual information that robot policies rely on, so the 75x-compressed training data does not reduce downstream task accuracy.

Editorial extensions

If this is right

  • If Robo-DM's compression holds, the 8.9 TB Open-X-Embodiment dataset could be distributed at a fraction of its current cloud egress cost, making it feasible for more labs to download and train on it.
  • Data-loading no longer bottlenecks training when episodes are small enough to cache: the paper reports no slowdown at batch size 64 when fine-tuning Octo, so larger-scale training runs could rely on Robo-DM without pipeline stalls.
  • Lossy video at CRF 30 may be acceptable for many robot manipulation tasks, potentially letting the community store and share far more demonstration data per terabyte.
  • The format's ROS2 compatibility means existing collection stacks can adopt Robo-DM without rewriting their data pipeline.

Reading between the lines

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

  • The 15/15 result is a single task with 335 demonstrations; a fair reader might infer that lossy compression's safety margin is task-dependent, and tasks requiring fine visual details (small parts, deformables, text reading) could degrade earlier.
  • The paper's load-balancing heuristic is a heuristic, and its RAM-heavy caching could make Robo-DM less attractive for very large single episodes; a natural extension would be adaptive cache sizing or multi-GPU decoding, which the paper does not test.
  • Since Robo-DM is codec-agnostic, the same container could test GPU-accelerated decoders (NVDEC) or newer codecs such as AV2/H.266, potentially improving both compression and throughput beyond AV1/CRF30.
  • The Octo fine-tune compared lossy training against lossless validation; a direct comparison of task success (not just MSE) on a physically evaluated lossless-trained model is the missing experiment that would fully support the 'no accuracy loss' claim.
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

3 major / 7 minor

Summary. The paper proposes Robo-DM, a data management toolkit that stores robot episodes in a self-contained EBML/MKV-based container, supports lossless and lossy video compression, and uses mmap-based caching and load balancing to accelerate data loading. The authors compare file sizes and loading throughput against RLDS, HDF5, and LeRobot on several Open-X-Embodiment datasets, and present two learning case studies: fine-tuning Octo on RT-1 data and training an ICRT policy for a physical pick-and-place task. The main claims are large size reductions (up to 70x lossy and 3.5x lossless vs RLDS), faster loading than LeRobot, and no reduction in downstream task accuracy.

Significance. If the size and speed claims hold, Robo-DM is a practical contribution to robot dataset management: the open-source release, the self-contained format, and the benchmarks against RLDS and LeRobot are valuable to the community. The paper also honestly flags a limitation about RAM-cache throughput degradation at scale in Sec. V.A. However, the downstream-accuracy claim is not yet supported by the evidence, and the speed claim contains an unsupported '50x' figure. These issues must be resolved before the paper can be accepted.

major comments (3)
  1. [Abstract, Sec. V.B, Sec. V.C] The abstract claims that Robo-DM 'does not suffer reduction in downstream task accuracy,' but the evidence does not establish this. The physical ICRT experiment (Sec. V.C) reports 15/15 successes on one pick-and-place task with no lossless-trained baseline, no repeated seeds, and no task variation; a ceiling effect or redundant visual cues could mask compression-induced degradation. The only quantitative comparison, the Octo fine-tune (Sec. V.B), shows validation image-conditioned MSE increasing from 1.86 (lossless) to 1.91 (lossy), a 2.6% degradation, with no error bars or significance test. This is a measurable reduction, so the 'no reduction' phrasing is not supported. Please add a lossless ICRT baseline, multiple seeds with variance reporting, and revise the abstract and conclusion to a claim such as 'minimal degradation' that matches the data.
  2. [Abstract, Sec. V.A, Fig. 4] The abstract states that Robo-DM is 'up to 50x faster when decoding sequentially,' but the reported experiments in Fig. 4 and Sec. V.A show speedups over LeRobot of 33x, 20x, and 5x; no experiment in the manuscript supports 50x. Furthermore, the LeRobot baseline is run using the official sequential extraction example, whereas Robo-DM uses concurrent batch loading with mmap caching and load balancing, so the comparison may conflate algorithmic gains with different evaluation protocols. Please report the access pattern for each baseline explicitly, provide sequential-vs-sequential and batch-vs-batch comparisons, and either remove the 50x figure or support it with a matching experiment.
  3. [Sec. V.B] The RT-1 compression ratio is reported as 4.39x, but the numbers given in the same paragraph imply a different ratio: the original dataset is 111.06 GB and the Robo-DM dataset is 36.50 GB, which is a 3.04x reduction. Please correct this numerical inconsistency and verify all derived ratios in the manuscript.
minor comments (7)
  1. [Sec. I, Sec. II, Sec. IV.C] There are several textual issues: 'structrure' in Sec. I should be 'structure,' 'infrastructrue' in Sec. II should be 'infrastructure,' and the 'Listing ??' placeholder in Sec. IV.C must be filled in.
  2. [Table I] The table header is garbled ('Dataset Avg. Frames Original Robo-DM- # Image Streams Resolution per Episode RLDS HDF5 Lossless LeRobot Robo-DM'). Please reformat so each column label aligns with its data, and define what 'Lossless' refers to (presumably Robo-DM with lossless codecs).
  3. [Fig. 4, Fig. 5] Please add error bars or confidence intervals and state whether the bars represent standard deviation or standard error; currently it is impossible to assess the variability of the latency measurements.
  4. [Sec. V.A] The LeRobot baseline is described as 'We sequentially extract episodes suggested by the example instructions.' Please clarify whether this matches the official LeRobot video benchmark [44] and whether the throughput comparison uses the same batch size and decoder settings for both frameworks.
  5. [Sec. V.C] Figure 6 shows a single task configuration; please describe the randomization procedure (object positions, number of distinct placements) and report trial-by-trial outcomes or at least the distribution of successes.
  6. [References] References [40] and [41] contain 'Accessed: [Insert Date]' placeholders; please replace them with the actual access dates.
  7. [Abstract, Table I] In the abstract and Table I, 'A V1' should be 'AV1', and the Constant Rate Factor (CRF) parameter should be defined at first use in the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Robo-DM's size, speed, and accuracy claims are evaluated against external formats and external benchmarks, not derived from its own definitions or fitted parameters.

full rationale

The paper's central claims are benchmark-based comparisons against external formats (RLDS, LeRobot, HDF5) and external training workloads. The compression ratio is a direct file-size measurement, not a quantity defined in terms of its own prediction. The AV1 CRF 30 setting is explicitly adopted from the LeRobot video benchmark [44], an external source, rather than tuned to force Robo-DM's results. The Octo fine-tuning comparison reports validation MSE on the original lossless data, so it is an independent check rather than a re-labeled fit. The 15/15 physical ICRT result lacks a lossless-trained baseline and is statistically weak, but that is an empirical support problem, not circularity: nothing in the protocol makes the reported success equivalent by construction to the compression setting. The paper's self-citations ([7] for ICRT, [47] for the UR5 dataset) are used as tools and test data, not as justifications that Robo-DM's compression preserves accuracy. The acknowledged RAM-cache limitation in Sec. V.A is an honest boundary condition and does not reveal a definitional loop. No equation in the paper is fitted to data and then presented as a prediction, and no load-bearing premise is imported from a self-citation chain. Under the stated criteria, the appropriate finding is no significant circularity.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claims rest on the adequacy of lossy video compression and on fair baselines, not on fitted equations or new physical entities. The only hand-chosen parameter affecting the compression/accuracy tradeoff is CRF 30, adopted from LeRobot's benchmark. No circular derivation is present.

free parameters (1)
  • AV1 Constant Rate Factor (CRF) = 30
    Hand-chosen lossy video quality setting, taken from LeRobot's recommended benchmark [44]; it controls the compression level and the compression/accuracy tradeoff, but is not fitted to Robo-DM's results.
assumptions (4)
  • domain assumption Lossy AV1 compression at CRF 30 preserves task-relevant visual information for policy learning
    The central accuracy-preservation claim rests on this; evidence is one 15-trial physical task with no lossless baseline and a 2.6% validation MSE increase in Octo.
  • domain assumption EBML/MKV containers can represent heterogeneous, time-aligned robot streams without losing synchronization
    Design foundation of Robo-DM's self-contained format, relying on EBML (RFC 8794) and Matroska multiplexing; no formal proof of synchronization guarantees is given.
  • domain assumption mmap-based caching and latency-based load balancing are implemented without contention that compromises throughput
    The loading speedups depend on this mechanism; the paper itself notes high RAM usage and throughput degradation for large per-episode data at larger batch sizes.
  • domain assumption Baseline conversions and loader configurations are representative and fair
    Speed comparisons rely on official RLDS/LeRobot scripts and a hand-set HDF5 prefetch buffer; LeRobot conversion omits depth and some action streams, which may favor Robo-DM.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Robo-DM: Data Management For Large Robot Datasets." pith.science (2026). https://pith.science/paper/A5D7DI5Z

@misc{pith2026250515558,
  author       = {Pith},
  title        = {Pith review of: Robo-DM: Data Management For Large Robot Datasets},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A5D7DI5Z}},
  note         = {Machine review of arXiv:2505.15558}
}
read the original abstract

Recent results suggest that very large datasets of teleoperated robot demonstrations can be used to train transformer-based models that have the potential to generalize to new scenes, robots, and tasks. However, curating, distributing, and loading large datasets of robot trajectories, which typically consist of video, textual, and numerical modalities - including streams from multiple cameras - remains challenging. We propose Robo-DM, an efficient open-source cloud-based data management toolkit for collecting, sharing, and learning with robot data. With Robo-DM, robot datasets are stored in a self-contained format with Extensible Binary Meta Language (EBML). Robo-DM can significantly reduce the size of robot trajectory data, transfer costs, and data load time during training. Compared to the RLDS format used in OXE datasets, Robo-DM's compression saves space by up to 70x (lossy) and 3.5x (lossless). Robo-DM also accelerates data retrieval by load-balancing video decoding with memory-mapped decoding caches. Compared to LeRobot, a framework that also uses lossy video compression, Robo-DM is up to 50x faster when decoding sequentially. We physically evaluate a model trained by Robo-DM with lossy compression, a pick-and-place task, and In-Context Robot Transformer. Robo-DM uses 75x compression of the original dataset and does not suffer reduction in downstream task accuracy.

Figures

Figures reproduced from arXiv: 2505.15558 by the authors.

Figure 1
Figure 1. Robo-DM can streamline robot data collection, management, and [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. A File Structure Comparison of RLDS, LeRobot and Robo-DM All formats include metadata, storing descriptive information such as authors and dataset summary. (A) Reinforcement Learning Dataset (RLDS) stores episodes in partitions, where each partition is a Tensorflow Dataset Record file. All streams in episode data are compressed matrices that can be directly loaded and trained in Tensorflow. (B) LeRobot combines thre… view at source ↗
Figure 3
Figure 3. How Robo-DM stores an episode of robot data with vision, language and action data Robo-DM encodes vision, language and action data. For vision data, Robo-DM uses video or image compression; language and action data are serialized into bytes. All the bytes are encapsulated with an intake timestamp. Then Robo-DM multiplexes different streams of data into a self-describing EBML file format. Efficient Decoding Cache For… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Episode Per Second Throughput of Robo-DM on Three OXE datasets with Different Characteristics We compare Robo-DM with baseline data loading Methods RLDS, HDF5 and LeRobot. Complete episodes are loaded concurrently as a batch, and we record the average throughput with 2…
Figure 5
Figure 5. Figure 5: Concurrent Loading Latency with respect to Episode Size of Robo-DM We compare Robo-DM with baseline data loading Methods RLDS, HDF5 and LeRobot. Complete episodes are loaded concurrently as a batch, and we record the average latency of 200 batches with batch size 8 epi…
Figure 6
Figure 6. Figure 6: ICRT Physical Experiment Setup with Robo-DM We setup ICRT to pick up a stuffed toy tiger and place it into a black bowl with a Franka Emika robot arm. The Figure shows the view from the left camera and wrist camera used for training, for both the original dataset and r…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 46 canonical work pages

  1. [1]

    Ghosh, H

    Octo Model Team, D. Ghosh, H. Walke, K. Pertsch, K. Black, O. Mees, S. Dasari, J. Hejna, C. Xu, J. Luo, T. Kreiman, Y . Tan, et al., Octo: An open-source generalist robot policy, https://octo- models.github.io, 2023

    work page 2023

  2. [2]

    Some features in Open-X-Embodiment are omitted in the conversion

    LeRobot [18] We convert Open-X-Embodiment datasets in LeRobot datasets with the provided official script. Some features in Open-X-Embodiment are omitted in the conversion. We sequentially extract episodes suggested by the example instructions. 3) HDF5 [16] We use Robo-DM to convert Open-X-Embodiment datasets to HDF5 formats. Since one HDF5 file per trajec...

    work page

  3. [3]

    RT- 1: Robotics transformer for real-world control at scale,

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, C. Finn, K. Gopalakr- ishnan, K. Hausman, A. Herzog, J. Hsu, B. Zitkovich, et al., “RT- 1: Robotics transformer for real-world control at scale,” Robotics: Science and Systems (RSS) , 2023

    work page 2023

  4. [4]

    O. X.-E. Collaboration et al., Open X-Embodiment: Robotic learning datasets and RT-X models , 2024

    work page 2024

  5. [5]

    Openvla: An open-source vision-language-action model,

    M. Kim, K. Pertsch, S. Karamcheti, T. Xiao, A. Balakrishna, S. Nair, R. Rafailov, E. Foster, G. Lam, P. Sanketi, Q. Vuong, T. Kollar, et al., “Openvla: An open-source vision-language-action model,” Conference on Robot Learning (CoRL) , 2024

    work page 2024

  6. [6]

    Rt-2: Vision-language-action models transfer web knowledge to robotic control,

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, C. Finn, K. Gopalakrishnan, K. Hausman, A. Herzog, J. Hsu, B. Zitkovich, et al., “Rt-2: Vision-language-action models transfer web knowledge to robotic control,” in Conference on Robot Learning , PMLR, 2023, pp. 2165–2183

    work page 2023

  7. [7]

    In-context imitation learning via next-token prediction,

    L. Fu, H. Huang, G. Datta, L. Y . Chen, W. C. -H. Panitch, F. Liu, H. Li, and K. Goldberg, “In-context imitation learning via next-token prediction,” International Conference on Robotics and Automation , 2025

    work page 2025

  8. [8]

    Droid: A large-scale in-the-wild robot manipulation dataset,

    A. Khazatsky, K. Pertsch, S. Nair, A. Balakrishna, S. Dasari, S. Karamcheti, S. Nasiriany, M. K. Srirama, L. Y . Chen, et al., “Droid: A large-scale in-the-wild robot manipulation dataset,” in Proceedings of Robotics: Science and Systems , Delft, Netherlands, 2024

    work page 2024

Show all 52 references
  1. [9]

    Gemini: A family of highly capable multimodal models,

    Google, “Gemini: A family of highly capable multimodal models,” arXiv preprint arXiv:2312.11805 , 2023

    2023 arXiv

  2. [10]

    Gpt-4v(ision) system card,

    “Gpt-4v(ision) system card,” 2023

    2023

  3. [11]

    Flamingo: A visual language model for few- shot learning,

    J.-B. Alayrac, J. Donahue, P. Luc, A. Miech, I. Barr, Y . Hasson, K. Lenc, A. Mensch, K. Millican, M. Reynolds, R. Ring, E. Rutherford, et al. , “Flamingo: A visual language model for few- shot learning,” Advances in neural information processing systems , vol. 35, pp. 23 716–...

    2022

  4. [12]

    Visual instruction tuning,

    H. Liu, C. Li, Q. Wu, and Y . J. Lee, “Visual instruction tuning,” Advances in neural information processing systems , vol. 36, 2024

    2024

  5. [13]

    Llama: Open and efficient foundation language models,

    H. Touvron, T. Lavril, G. Izacard, X. Martinet, M. -A. Lachaux, T. Lacroix, B. Rozi`ere, N. Goyal, E. Hambro, F. Azhar, A. Rodriguez, A. Joulin, et al., “Llama: Open and efficient foundation language models,” arXiv preprint arXiv:2302.13971 , 2023

    2023 arXiv

  6. [14]

    Palm-e: An embodied multimodal language model,

    D. Driess, F. Xia, M. S. M. Sajjadi, C. Lynch, A. Chowdhery, B. Ichter, A. Wahid, J. Tompson, Q. Vuong, T. Yu, W. Huang, Y . Chebotar, et al. , “Palm-e: An embodied multimodal language model,” in International Conference on Machine Learning , PMLR, 2023, pp. 8469–8488

    2023

  7. [15]

    Big data: A review,

    S. Sagiroglu and D. Sinanc, “Big data: A review,” in 2013 interna- tional conference on collaboration technologies and systems (CTS) , IEEE, 2013, pp. 42–47

    2013

  8. [16]

    Gpt-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat, et al., “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774 , 2023

    2023 arXiv

  9. [17]

    Rlds: An ecosystem to generate, share and use datasets in reinforcement learning,

    S. Ramos, S. Girgin, L. Hussenot, D. Vincent, H. Yakubovich, D. Toyama, A. Gergely, P. Stanczyk, R. Marinier, J. Harmsen, O. Pietquin, and N. Momchev, “Rlds: An ecosystem to generate, share and use datasets in reinforcement learning,” arXiv preprint arXiv:2111.02767, 2021

    2021 arXiv

  10. [18]

    The HDF Group, Hierarchical Data Format, version 5 , 1997-2024

    1997

  11. [19]

    Apache Arrow, https://arrow.apache.org/, 2024-09-14

    2024

  12. [20]

    Cadene, S

    R. Cadene, S. Alibert, A. Soare, Q. Gallouedec, and T. Wolf, Lerobot: Making ai for robotics more accessible with end-to-end learning , https://github.com/huggingface/lerobot, 2024

    2024

  13. [21]

    Scalable deep reinforcement learning for vision-based robotic manipulation,

    D. Kalashnikov, A. Irpan, P. Pastor, J. Ibarz, A. Herzog, E. Jang, D. Quillen, E. Holly, M. Kalakrishnan, V . Vanhoucke, and S. Levine, “Scalable deep reinforcement learning for vision-based robotic manipulation,” in Conference on robot learning , PMLR, 2018, pp. 651–673

    2018

  14. [22]

    com / huggingface/safetensors, Accessed: 2024-09-14

    HuggingFace SafeTensors , https : / / github . com / huggingface/safetensors, Accessed: 2024-09-14

    2024

  15. [23]

    ROS: An open-source robot operating system,

    M. Quigley, B. Gerkey, K. Conley, J. Faust, T. Foote, J. Leibs, R. Wheeler, and A. Ng, “ROS: An open-source robot operating system,” in ICRA workshop on open source software , vol. 3, 2009

    2009

  16. [24]

    Bridgedata v2: A dataset for robot learning at scale,

    H. Walke, K. Black, A. Lee, M. J. Kim, M. Du, C. Zheng, T. Zhao, P. Hansen-Estruch, Q. Vuong, A. He, V . Myers, K. Fang, et al. , “Bridgedata v2: A dataset for robot learning at scale,” in Conference on Robot Learning , PMLR, 2023, pp. 1723–1736

    2023

  17. [25]

    Fog robotics: An introduction,

    S. L. K. C. Gudi, S. Ojha, B. Johnston, J. Clark, and M. -A. Williams, “Fog robotics: An introduction,” in IEEE/RSJ International Conference on Intelligent Robots and Systems , 2017

    2017

  18. [26]

    TensorFlow Datasets, a collection of ready-to-use datasets , https: //www.tensorflow.org/datasets

  19. [27]

    Fog robotics algorithms for distributed motion planning using lambda serverless computing,

    J. Ichnowski, W. Lee, V . Murta, S. Paradis, R. Alterovitz, J. E. Gonzalez, I. Stoica, and K. Goldberg, “Fog robotics algorithms for distributed motion planning using lambda serverless computing,” in Proc. IEEE Int. Conf. Robotics and Automation (ICRA) , 2020, pp. 4232–4238

    2020

  20. [28]

    A fog robotics approach to deep robot learning: Application to object recognition and grasp planning in surface decluttering,

    A. K. Tanwani, N. Mor, J. Kubiatowicz, J. E. Gonzalez, and K. Goldberg, “A fog robotics approach to deep robot learning: Application to object recognition and grasp planning in surface decluttering,” in Proc. IEEE Int. Conf. Robotics and Automation (ICRA), IEEE, 2019, pp. 4559–4566

    2019

  21. [29]

    Fog robotics for efficient, fluent and robust human-robot interaction,

    S. L. K. C. Gudi, S. Ojha, B. Johnston, J. Clark, and M. -A. Williams, “Fog robotics for efficient, fluent and robust human-robot interaction,” in 2018 IEEE 17th International Symposium on Network Computing and Applications (NCA) , IEEE, 2018, pp. 1–5

    2018

  22. [30]

    A fog robotic system for dynamic visual servoing,

    N. Tian, A. K. Tanwani, J. Chen, M. Ma, R. Zhang, B. Huang, K. Goldberg, and S. Sojoudi, “A fog robotic system for dynamic visual servoing,” in 2019 International Conference on Robotics and Automation (ICRA), IEEE, 2019, pp. 1982–1988

    2019

  23. [31]

    FogROS2-SGC: A ROS2 cloud robotics platform for secure global connectivity,

    K. Chen, R. Hoque, K. Dharmarajan, E. Llontop, S. O. Adebola, J. Ichnowski, J. D. Kubiatowicz, and K. Goldberg, “FogROS2-SGC: A ROS2 cloud robotics platform for secure global connectivity,” 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 1–8, 2023

    2023

  24. [32]

    FogROS: An adaptive framework for automating fog robotics deployment,

    K. E. Chen, Y . Liang, N. Jha, J. Ichnowski, M. Danielczuk, J. Gonzalez, J. Kubiatowicz, and K. Goldberg, “FogROS: An adaptive framework for automating fog robotics deployment,” in 2021 IEEE 17th International Conference on Automation Science and Engineering (CASE) , IEEE, 202...

    2021

  25. [33]

    FogROS2-Config: A toolkit for choosing server configuration for cloud robotics,

    K. Chen, K. Hari, R. Khare, C. Le, T. Chung, J. Drake, S. Adebloa, J. Ichnowski, J. Kubiatowicz, and K. Goldberg, “FogROS2-Config: A toolkit for choosing server configuration for cloud robotics,” Proc. IEEE Int. Conf. Robotics and Automation (ICRA) , 2024

    2024

  26. [34]

    FogROS2-LS: A location-independent fog robotics framework for latency sensitive ROS2 applications,

    K. Chen, M. Wang, M. Gualtieri, N. Tian, C. Juette, L. Ren, J. Kubia- towicz, and K. Goldberg, “FogROS2-LS: A location-independent fog robotics framework for latency sensitive ROS2 applications,” Proc. IEEE Int. Conf. Robotics and Automation (ICRA) , 2024

    2024

  27. [35]

    Foxglove Technologies Inc, Foxglove, https://foxglove.dev

  28. [36]

    EBML is a versatile and extensible markup language that combines the flexibility of Extensible Meta Language (XML) with the efficiency of binary encoding

    for data structuring. EBML is a versatile and extensible markup language that combines the flexibility of Extensible Meta Language (XML) with the efficiency of binary encoding. It organizes binary data elements in a hierarchical structure similar to XML, allowing for nested el...

  29. [37]

    Rviz: A toolkit for real domain data visualization,

    H. R. Kam, S. -H. Lee, T. Park, and C. -H. Kim, “Rviz: A toolkit for real domain data visualization,” Telecommunication Systems, vol. 60, no. 2, pp. 337–345, 2015

    2015

  30. [38]

    Extensible binary meta language,

    S. Lhomme, D. Rice, and M. Bunkus, “Extensible binary meta language,” RFC Editor, RFC 8794, Jul. 2020

    2020

  31. [39]

    Matroska Video Container , https://www.matroska.org/ index.html, Accessed: 2024-09-14

    2024

  32. [40]

    Advanced video coding for generic audiovisual services,

    ITU-T, “Advanced video coding for generic audiovisual services,” International Telecommunication Union, Geneva, Switzerland, Rec- ommendation H.264, 2003

    2003

  33. [41]

    High efficiency video coding,

    ITU-T, “High efficiency video coding,” International Telecommuni- cation Union, Geneva, Switzerland, Recommendation H.265, 2023, Version 9

    2023

  34. [42]

    Alliance for Open Media, Av1 bitstream & decoding process specification, https://aomediacodec.github.io/av1- spec/, Accessed: [Insert Date], 2019

    2019

  35. [43]

    Library of Congress, Ff video codec 1, version 0, 1 and 3 , https: / / www . loc . gov / preservation / digital / formats / fdd/fdd000341.shtml, Accessed: [Insert Date], 2024

    2024

  36. [44]

    Ffv1 video coding format versions 0, 1, and 3,

    M. Niedermayer, D. Rice, and J. Martinez, “Ffv1 video coding format versions 0, 1, and 3,” RFC Editor, RFC 9043, Aug. 2021

    2021

  37. [45]

    Linux mmap(2) Manual , https://man7.org/linux/man- pages/man2/mmap.2.html, Accessed: 2024-09-14

    2024

  38. [46]

    com / huggingface / lerobot / tree / main / benchmarks / video, Accessed: 2024-09-13

    LeRobot Video Benchmark , https : / / github . com / huggingface / lerobot / tree / main / benchmarks / video, Accessed: 2024-09-13

    2024

  39. [47]

    Multi-stage cable routing through hierarchical imitation learning,

    J. Luo, C. Xu, X. Geng, G. Feng, K. Fang, L. Tan, S. Schaal, and S. Levine, “Multi-stage cable routing through hierarchical imitation learning,” IEEE Transactions on Robotics , 2024

    2024

  40. [48]

    The surprising effectiveness of representation learning for visual imitation,

    J. Pari, N. M. Shafiullah, S. P. Arunachalam, and L. Pinto, “The surprising effectiveness of representation learning for visual imitation,” Robotics: Science and Systems , 2018

    2018

  41. [49]

    L. Y . Chen, S. Adebola, and K. Goldberg, Berkeley UR5 demonstra- tion dataset, https://sites.google.com/view/berkeley-ur5/home

  42. [50]

    Pyav: Pythonic bindings for FFmpeg’s libraries , https : / / github.com/PyAV-Org/PyAV, Accessed: 2024-09-14

    2024

  43. [51]

    Decord: An efficient video loader for deep learning with smart shuffling that’s super easy to digest , https://github.com/ dmlc/decord, Accessed: 2024-09-14

    2024

  44. [52]

    NVIDIA Video Codec SDK , https : / / developer . nvidia . com/video-codec-sdk , Accessed: 2024-09-14

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.