Pith. sign in

REVIEW 3 major objections 3 minor 1 cited by

Human2LocoMan: Learning Versatile Quadrupedal Manipulation with Human Pretraining

T0 review · 3 major / 3 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Pretraining a quadruped-manipulation policy on human demonstrations, then finetuning it on a small robot dataset, raises real-world success rates by 38.6% on average and 82.7% on out-of-distribution objects.

desk verdict A solid cross-embodiment IL system with real robot gains from human pretraining, but the headline OOD numbers are inflated because the 'OOD' objects were already in the human pretraining data. read the letter →

arxiv 2506.16475 v2 pith:FMFDLHK4 submitted 2025-06-19 cs.RO cs.AIcs.LG

Yaru Niu , Yunzhe Zhang , Mingyang Yu , Changyi Lin , Chenhao Li , Yikai Wang , Yuxiang Yang , Wenhao Yu
show 6 more authors
Tingnan Zhang Zhenzhen Li Jonathan Francis Bingqing Chen Jie Tan Ding Zhao
This is my paper · ORCID
classification cs.ROcs.AIcs.LG
keywords quadrupedalmanipulationimitationlearningcross-embodimenttransferhumanpretrainingteleoperationmodulartransformerpolicyloco-manipulationout-of-distributiongeneralization
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

This paper tries to establish that a quadrupedal robot with leg-mounted manipulators can learn household manipulation skills by first pretraining a policy on human demonstrations and then finetuning it on a small set of robot demonstrations. To make that possible, the authors build a teleoperation and data-collection pipeline that records human head, wrist, and grip motions and maps them to the robot's torso, end effectors, and grippers in a common coordinate frame, along with a Modularized Cross-embodiment Transformer (MXT) whose shared trunk is flanked by embodiment-specific tokenizers and detokenizers. On six real-world tasks spanning unimanual and bimanual, prehensile and non-prehensile, tool-use and long-horizon manipulation, the system reports large gains: pretraining with human data adds 38.6% success overall and 82.7% under out-of-distribution settings, and a pretrained model matches or beats baselines using only half the robot data. If correct, this is evidence that easy-to-collect human demonstrations can substitute for expensive robot demonstrations even when the human and robot bodies look very different.

What carries the argument

The Modularized Cross-embodiment Transformer (MXT) — a transformer policy with a shared encoder-decoder trunk and per-embodiment, per-modality tokenizers and detokenizers — is the mechanism that carries the argument. It makes the policy consume a fixed sequence of tokens in which each semantic modality (main image, wrist image, body pose, end-effector pose, gripper action) is encoded and decoded separately, so the shared trunk sees aligned modality structure even when the human and quadruped embodiments differ; missing modalities are handled by masks. The teleoperation mapping in Eq. (1) plays the supporting role of turning human head and wrist motions into target torso and end-effector poses in a unified frame, so the human and robot datasets are spatially commensurable before training begins.

What would settle it

Log the commanded target poses from Eq. (1) against the robot's actual torso and end-effector poses throughout teleoperation on every task, and compute the per-axis pose tracking error; then retrain the finetuned policy on only those demonstrations whose tracking error stays below a threshold. If success rates do not drop when poorly tracked demonstrations are removed, or if tracking error is large even in successful rollouts, the human-pretraining gain is not explained by faithful demonstrations.

Watch

Extended reading notes

Core claim

The central claim is that positive cross-embodiment transfer from humans to a quadruped works when both embodiments' observations and actions are decomposed into aligned modalities (images, body pose, end-effector pose, gripper state) and processed by separate tokenizers and detokenizers around a shared transformer trunk. The human data are collected first, the MXT policy is pretrained on them with behavior cloning, then the tokenizers and detokenizers are reinitialized and the trunk is finetuned on LocoMan trajectories recorded through the same teleoperation interface. Across six tasks the pretrained policy reaches above 79% success on in-distribution settings and substantially higher out-of-distribution success than without pretraining: an average success-rate gain of 38.6% overall (41.9% vs the stronger imitation baseline) and 82.7% under OOD conditions, with gains concentrated in precise, long-horizon substeps such as scooping and pouring. The paper interprets this as the trunk learning modality-level manipulation structure from human data that transfers to a very different body.

Load-bearing premise

The whole result rests on the assumption that the teleoperation mapping and the whole-body controller actually deliver the target torso and end-effector poses with enough fidelity that the recorded robot trajectories are valid demonstrations; tracking error is never measured, so if the robot lags or distorts the human commands, both the robot dataset and the measured transfer gains are called into question.

Editorial extensions

If this is right

  • Human demonstrations can replace about half of the robot demonstrations without hurting success, cutting the time and hardware cost of data collection for quadrupedal manipulation.
  • A single pretrained trunk can be finetuned into different manipulation modes (unimanual and bimanual) of the same robot, so pretraining cost is amortized across tasks.
  • Out-of-distribution generalization — new object shapes, colors, and placements — is where human pretraining helps most, not just average in-distribution performance.
  • The modular modality split, not merely the shared transformer trunk, is what preserves transfer; aggregating modalities into single tokenizers removes most of the benefit.
  • Human pretraining appears to matter most for precise substeps inside long-horizon tasks, such as localizing and scooping litter or pouring a ball, suggesting the transferred knowledge is about accuracy of coordinated motion, not just task semantics.

Reading between the lines

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

  • If the transfer effect is as general as the paper's mechanism suggests, the same XR-captured human data could pretrain a trunk for other legged or wheeled manipulator platforms, provided their observations and actions are expressed in the same modality-aligned format; this is a testable extension the paper does not run.
  • The result implies that kinematic similarity between demonstrator and robot may not be necessary for imitation pretraining, only structured alignment of modalities; that is an editorial inference because the paper only demonstrates one robot body.
  • Because human data are cheaper and faster to collect (over 200 trajectories in 30 minutes, compared with over 50 robot trajectories), the measured gains suggest the bottleneck in robot imitation learning could shift from data volume to the quality of the teleoperation-to-robot mapping.
  • A natural next experiment, which the paper lists as future work, is simultaneous co-training on human and robot data in a multi-task setting; the sequential pretrain-finetune design leaves open whether mixing data during training would preserve or amplify the transfer.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. Human2LocoMan introduces an XR-based teleoperation and data-collection pipeline that unifies human and LocoMan observations and actions, together with a modularized cross-embodiment Transformer (MXT) that is pretrained on human demonstrations and then finetuned on small robot datasets. The paper reports real-robot success rates on six household tasks in unimanual and bimanual modes, and claims that human pretraining contributes a 38.6% success-rate improvement overall and 82.7% under OOD settings, with larger gains over the HIT baseline. The paper also releases code, hardware, and data.

Significance. If the central transfer claim held, this would be a meaningful contribution to cross-embodiment imitation learning for quadrupedal manipulation, combining a practical data-collection interface with an open-source dataset and policy architecture. The within-distribution comparisons in Table II do show consistent positive deltas for MXT-Pretrained over MXT-Scratch on several tasks, and the modular MXT design outperforms the HPT-style aggregated baseline in the ablation. The paper is also commendable for open-sourcing code, hardware, and data and for evaluating on real hardware rather than only in simulation. However, the headline OOD generalization claim is currently confounded by pretraining-data leakage for the toy-collection tasks, and the aggregate percentages in the abstract do not reproduce from the reported table under standard aggregation rules.

major comments (3)
  1. [IV-A.1, IV-B, Table II] For the Toy Collection tasks, Section IV-A.1 states that "we use 10 objects for robot finetuning and all objects for human pretraining and real-robot evaluation," and Figure 4 labels the held-out test items as OOD objects. For MXT-Pretrained, these objects are therefore not out-of-distribution: their appearance and affordances were seen during human-data pretraining. Since the OOD rows are the strongest evidence for embodiment-agnostic transfer, the claimed 82.7% OOD gain conflates object familiarity acquired during pretraining with cross-embodiment skill transfer. Please either exclude the OOD evaluation objects from human pretraining, or rename and re-analyze the condition as "novel to robot finetuning but seen in human pretraining," and adjust the central OOD claims accordingly.
  2. [IV-B, Table II] The aggregated improvements "38.6% overall and 82.7% under OOD" for MXT-Pretrained over MXT-Scratch, and "41.9% overall and 79.7% under OOD" over HIT, are not reproducible from Table II under standard aggregation. For example, averaging the per-task OOD success-rate deltas between MXT-Pretrained (larger) and MXT-Scratch (larger) gives 31.95 percentage points, and including both dataset sizes gives about 33.3 percentage points; I could not reproduce 82.7 with natural weightings of the 24 ID and 12 OOD trials. Please state the exact aggregation rule (which rows, which baseline, absolute or relative changes) and verify the headline numbers, since the abstract and introduction lead with these quantities.
  3. [Figure 5 caption and Table II] The caption of Figure 5 states that results are based on 36 trials comprising 24 OOD and 12 ID trials, whereas Section IV-A.6 and Table II state that ID results use 24 trials and OOD results use 12 trials. Please correct this inconsistency; if OOD trials were actually 24, the reported OOD percentages and the statistical strength of the OOD claims would need to be recomputed.
minor comments (3)
  1. [III-B, Eq. (1)] The notation in Eq. (1) refers to "xgripper max" and "xgripper min" as gripper angles, but the variables elsewhere are denoted by θ; please unify the notation.
  2. [IV-A.6, Table II] Success rates with 12 OOD trials distinguish only a handful of successes per cell; several reported differences correspond to 1–4 successful rollouts. Including raw counts or confidence intervals would help the reader calibrate the OOD comparisons.
  3. [III-B] The whole-body controller's tracking error is not quantified; since the teleoperation pipeline's validity depends on the controller following the mapped target poses, reporting measured end-effector and torso tracking error would strengthen the data-collection claims.

Circularity Check

1 steps flagged · score 6.0 of 10

OOD transfer claim is confounded: the 'OOD' objects are inside the human pretraining set, so the 82.7% OOD gain largely measures in-distribution object familiarity rather than cross-embodiment generalization.

  1. fitted input called prediction [Section IV-A.1 (Tasks), TC-Uni and TC-Bi descriptions; relied on in Section IV-B(3) and the Abstract OOD claims]
    "As shown in Figure 4, we use 10 objects for robot finetuning and all objects for human pretraining and real-robot evaluation."

    The OOD success-rate gain attributed to human pretraining is computed on objects that are explicitly included in the human pretraining set: 'all objects for human pretraining and real-robot evaluation.' Thus the 'OOD' test set is not out-of-distribution for the pretrained MXT; by construction it is a subset of the pretraining distribution. The claimed 82.7% OOD improvement is therefore a comparison between a model that has seen the test objects through human data and a scratch model that has not. The 'OOD generalization' result is forced by this experimental split rather than being a prediction of behavior on genuinely novel objects, and the same overlap inflates the aggregate 38.6% overall improvement.

full rationale

The central empirical pipeline is not circular: MXT pretraining and finetuning are evaluated by real-robot rollouts, and the success-rate gains are not derived from the loss or the data format. The whole-body controller adapted from [14] is a platform dependency, not a load-bearing self-citation; the transfer claim is not justified merely by citing [14]. The one significant circular/confounded element is the OOD evaluation: for the Toy Collection tasks, the objects used to compute OOD success are explicitly part of the human pretraining set, so the headline '82.7% under OOD settings' is an in-distribution measurement for the pretrained model. This does not fully destroy the cross-embodiment story, because independent ID gains remain (e.g., TC-Uni ID 91.7 vs 70.8 and Scoop ID 87.5 vs 62.5 for smaller data), so positive transfer has non-circular support. However, the OOD generalization claim is partially circular and should be interpreted as object familiarity plus transfer rather than pure embodiment-agnostic generalization.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central empirical result is supported by real-robot rollouts rather than derived from the model, so the main hidden costs are the teleoperation mapping constants, task-specific dropout rates, unstated training-subset selection criteria, and assumptions about the whole-body controller and the meaning of OOD relative to human pretraining data.

free parameters (3)
  • Motion scaling factors (alpha_torso, alpha_r-eef, alpha_l-eef) = not reported
    Eq. 1 uses these constants to map human head and wrist displacements to LocoMan torso and end-effector targets. They determine the teleoperation gain and reach; their values are not given, so data collection cannot be exactly reproduced and sensitivity is unexamined.
  • Pretraining dropout rates = 0.4 for most tasks, 0.5 for scooping
    Appendix A states these were chosen to improve finetuning performance. They are task-specific tuning choices that influence the measured transfer gain.
  • Finetuning data subsets = 20/40 trajectories for TC-Uni, 30/60 for TC-Bi, and similar splits for other tasks
    Table III reports up to 150 collected robot trajectories per task, while Table II footnote reports using only 20 to 80 for training. The criteria for selecting these subsets are not described, which could affect reported success rates.
assumptions (4)
  • domain assumption The LocoMan whole-body controller accurately tracks the commanded torso and end-effector target poses during teleoperation.
    Section III-B sends target poses p_t to the controller adapted from [14]; tracking error is not quantified. If tracking is poor, the recorded robot actions are not valid expert demonstrations and the transfer result is contaminated.
  • domain assumption Human head and wrist motion, after linear scaling in Eq. 1, is a useful prior for LocoMan's torso and end-effector motion.
    The entire pretraining stage relies on this mapping producing demonstrations that share structure with the robot's action space. It is only tested indirectly through downstream task success.
  • ad hoc to paper OOD evaluation objects may be considered out-of-distribution when they appear in the human pretraining set.
    Section IV-A states all objects are used in human pretraining while only 10 are used in robot finetuning; the paper's OOD claims are therefore with respect to robot data only, not to all training data.
  • domain assumption Success rates from 24 in-distribution and 12 OOD rollouts are stable enough to rank methods.
    Section IV-A-6 defines the evaluation; no confidence intervals or significance tests are given, so the assumption that observed gaps reflect underlying policy quality is not directly supported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Human2LocoMan: Learning Versatile Quadrupedal Manipulation with Human Pretraining." pith.science (2026). https://pith.science/paper/FMFDLHK4

@misc{pith2026250616475,
  author       = {Pith},
  title        = {Pith review of: Human2LocoMan: Learning Versatile Quadrupedal Manipulation with Human Pretraining},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FMFDLHK4}},
  note         = {Machine review of arXiv:2506.16475}
}
read the original abstract

Quadrupedal robots have demonstrated impressive locomotion capabilities in complex environments, but equipping them with autonomous versatile manipulation skills in a scalable way remains a significant challenge. In this work, we introduce a cross-embodiment imitation learning system for quadrupedal manipulation, leveraging data collected from both humans and LocoMan, a quadruped equipped with multiple manipulation modes. Specifically, we develop a teleoperation and data collection pipeline, which unifies and modularizes the observation and action spaces of the human and the robot. To effectively leverage the collected data, we propose an efficient modularized architecture that supports co-training and pretraining on structured modality-aligned data across different embodiments. Additionally, we construct the first manipulation dataset for the LocoMan robot, covering various household tasks in both unimanual and bimanual modes, supplemented by a corresponding human dataset. We validate our system on six real-world manipulation tasks, where it achieves an average success rate improvement of 41.9% overall and 79.7% under out-of-distribution (OOD) settings compared to the baseline. Pretraining with human data contributes a 38.6% success rate improvement overall and 82.7% under OOD settings, enabling consistently better performance with only half the amount of robot data. Our code, hardware, and data are open-sourced at: https://human2bots.github.io.

Figures

Figures reproduced from arXiv: 2506.16475 by the authors.

Figure 1
Figure 1. Human2LocoMan provides a unified framework for collecting human demonstrations and teleoperated robot whole￾body motions, along with cross-embodiment policy learning for quadrupedal manipulation. Human data is used for model pretraining, while robot data is leveraged for policy finetuning. Human2LocoMan achieves positive transfer from human to quadrupedal embodiments, facilitating versatile manipulation skills for u… view at source ↗
Figure 2
Figure 2. Human2LocoMan framework. Our system uses an XR headset for data collection, capturing egocentric human data and teleoperated robot data, all mapped to a unified coordinate frame. The dataset consists of aligned vision, proprioception, and actions from the human and the robot. We adopt a two-stage training process: the modularized cross-embodiment model is first pretrained on easy-to-collect human data, and then fine… view at source ↗
Figure 3
Figure 3. Modularized Cross-embodiment Transformer (MXT) architecture. The inputs are organized as a list of modalities and encoded each by a separate tokenizer into a fixed number of tokens. The Transformer trunk handles decision making by consuming the concatenated encoded tokens and producing a fixed number of raw output tokens. Each of the detokenizers at the end decodes a fixed subset of the output tokens into a modality… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Rollouts of the MXT policy and the objects used across manipulation tasks in our experiments. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Ablation study on unimanual and bimanual toy collection. We compare MXT, its ablation MXT-Agg, and baseline [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Substep success rate. The success rate for some substep [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Best validation loss of our method and HIT on [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Best validation loss of our method and HPT on the [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Being-H0: Vision-Language-Action Pretraining from Large-Scale Human Videos

    cs.CV 2025-07 conditional novelty 6.0 of 10

    A dexterous VLA pretrained on a 2.5M-instance human hand motion dataset transfers skills to a real robot hand, outperforming baselines in manipulation tasks.

Reference graph

Works this paper leans on

84 extracted references · 29 canonical work pages · cited by 1 Pith paper

  1. [1]

    Dtc: Deep tracking control

    Fabian Jenelten, Junzhe He, Farbod Farshidian, and Marco Hutter. Dtc: Deep tracking control. Science Robotics, 9(86):eadh5401, 2024

    2024

  2. [2]

    Learning quadrupedal locomotion on deformable terrain

    Suyoung Choi, Gwanghyeon Ji, Jeongsoo Park, Hyeongjun Kim, Juhyeok Mun, Jeong Hyun Lee, and Jemin Hwangbo. Learning quadrupedal locomotion on deformable terrain. Science Robotics , 8(74):eade2256, 2023

    2023

  3. [3]

    Learning quadrupedal locomotion over challenging terrain

    Joonho Lee, Jemin Hwangbo, Lorenz Wellhausen, Vladlen Koltun, and Marco Hutter. Learning quadrupedal locomotion over challenging terrain. Science robotics, 5 (47):eabc5986, 2020

    2020

  4. [4]

    Neural volumetric memory for visual locomotion control

    Ruihan Yang, Ge Yang, and Xiaolong Wang. Neural volumetric memory for visual locomotion control. In CVPR 2023, 2023

    2023

  5. [5]

    Agile continuous jumping in discontinuous terrains

    Yuxiang Yang, Guanya Shi, Changyi Lin, Xiangyun Meng, Rosario Scalise, Mateo Guaman Castro, Wenhao Yu, Tingnan Zhang, Ding Zhao, Jie Tan, et al. Agile continuous jumping in discontinuous terrains. arXiv preprint arXiv:2409.10923, 2024

    arXiv 2024

  6. [6]

    Rma: Rapid motor adaptation for legged robots

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

    arXiv 2021

  7. [7]

    Multimodality robotic systems: Integrated com- bined legged-aerial mobility for subterranean search-and- rescue

    Bj ¨orn Lindqvist, Samuel Karlsson, Anton Koval, Il- ias Tevetzidis, Jakub Halu ˇska, Christoforos Kanellakis, Ali-akbar Agha-mohammadi, and George Nikolakopou- los. Multimodality robotic systems: Integrated com- bined legged-aerial mobility for subterranean search-and- rescue. Robotics and Autonomous Systems , 2022

    work page 2022

  8. [8]

    Deep whole-body control: learning a unified policy for manipu- lation and locomotion

    Zipeng Fu, Xuxin Cheng, and Deepak Pathak. Deep whole-body control: learning a unified policy for manipu- lation and locomotion. In Conference on Robot Learning, pages 138–149. PMLR, 2023

    work page 2023

Show all 84 references
  1. [9]

    Helpful doggybot: Open-world object fetching using legged robots and vision-language models

    Qi Wu, Zipeng Fu, Xuxin Cheng, Xiaolong Wang, and Chelsea Finn. Helpful doggybot: Open-world object fetching using legged robots and vision-language models. arXiv preprint arXiv:2410.00231 , 2024

    2024 arXiv

  2. [10]

    Umi on legs: Making manipulation policies mobile with manipulation-centric whole-body controllers

    Huy Ha, Yihuai Gao, Zipeng Fu, Jie Tan, and Shuran Song. Umi on legs: Making manipulation policies mobile with manipulation-centric whole-body controllers. arXiv preprint arXiv:2407.10353, 2024

    2024 arXiv

  3. [11]

    Legato: Cross-embodiment imitation using a grasping tool

    Mingyo Seo, H Andy Park, Shenli Yuan, Yuke Zhu, and Luis Sentis. Legato: Cross-embodiment imitation using a grasping tool. arXiv preprint arXiv:2411.03682, 2024

    2024 arXiv

  4. [12]

    Visual manipu- lation with legs

    Xialin He, Chengjing Yuan, Wenxuan Zhou, Ruihan Yang, David Held, and Xiaolong Wang. Visual manipu- lation with legs. arXiv preprint arXiv:2410.11345, 2024

    2024 arXiv

  5. [13]

    Wildlma: Long horizon loco-manipulation in the wild

    Ri-Zhao Qiu, Yuchen Song, Xuanbin Peng, Sai Aneesh Suryadevara, Ge Yang, Minghuan Liu, Mazeyu Ji, Chengzhe Jia, Ruihan Yang, Xueyan Zou, et al. Wildlma: Long horizon loco-manipulation in the wild. arXiv preprint arXiv:2411.15131, 2024

    2024 arXiv

  6. [14]

    Locoman: Advancing versatile quadrupedal dexterity with lightweight loco-manipulators

    Changyi Lin, Xingyu Liu, Yuxiang Yang, Yaru Niu, Wenhao Yu, Tingnan Zhang, Jie Tan, Byron Boots, and Ding Zhao. Locoman: Advancing versatile quadrupedal dexterity with lightweight loco-manipulators. arXiv preprint arXiv:2403.18197, 2024

    2024 arXiv

  7. [15]

    Learning from demonstration

    Stefan Schaal. Learning from demonstration. Advances in neural information processing systems , 9, 1996

    1996

  8. [16]

    Viola: Imitation learning for vision-based manipulation with object proposal priors

    Yifeng Zhu, Abhishek Joshi, Peter Stone, and Yuke Zhu. Viola: Imitation learning for vision-based manipulation with object proposal priors. In Conference on Robot Learning, pages 1199–1210. PMLR, 2023

    2023

  9. [17]

    Mimicplay: Long-horizon imitation learning by watching human play

    Chen Wang, Linxi Fan, Jiankai Sun, Ruohan Zhang, Li Fei-Fei, Danfei Xu, Yuke Zhu, and Anima Anand- kumar. Mimicplay: Long-horizon imitation learning by watching human play. arXiv preprint arXiv:2302.12422, 2023

    2023 arXiv

  10. [18]

    Learning fine-grained bimanual manipulation with low-cost hardware

    Tony Z Zhao, Vikash Kumar, Sergey Levine, and Chelsea Finn. Learning fine-grained bimanual manipulation with low-cost hardware. arXiv preprint arXiv:2304.13705 , 2023

    2023 arXiv

  11. [19]

    Gello: A general, low-cost, and intuitive teleoperation framework for robot manipulators

    Philipp Wu, Yide Shentu, Zhongke Yi, Xingyu Lin, and Pieter Abbeel. Gello: A general, low-cost, and intuitive teleoperation framework for robot manipulators. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 12156–12163. IEEE, 2024

    2024

  12. [20]

    Humanplus: Humanoid shad- owing and imitation from humans

    Zipeng Fu, Qingqing Zhao, Qi Wu, Gordon Wet- zstein, and Chelsea Finn. Humanplus: Humanoid shad- owing and imitation from humans. arXiv preprint arXiv:2406.10454, 2024

    2024 arXiv

  13. [21]

    Open-television: Teleoperation with immersive active visual feedback

    Xuxin Cheng, Jialong Li, Shiqi Yang, Ge Yang, and Xiaolong Wang. Open-television: Teleoperation with immersive active visual feedback. arXiv preprint arXiv:2407.01512, 2024

    2024 arXiv

  14. [22]

    Omnih2o: Universal and dexterous human- to-humanoid whole-body teleoperation and learning

    Tairan He, Zhengyi Luo, Xialin He, Wenli Xiao, Chong Zhang, Weinan Zhang, Kris Kitani, Changliu Liu, and Guanya Shi. Omnih2o: Universal and dexterous human- to-humanoid whole-body teleoperation and learning. arXiv preprint arXiv:2406.08858 , 2024

    2024 arXiv

  15. [23]

    Ace: A cross-platform visual-exoskeletons system for low-cost dexterous teleoperation

    Shiqi Yang, Minghuan Liu, Yuzhe Qin, Runyu Ding, Jialong Li, Xuxin Cheng, Ruihan Yang, Sha Yi, and Xi- aolong Wang. Ace: A cross-platform visual-exoskeletons system for low-cost dexterous teleoperation. arXiv preprint arXiv:2408.11805, 2024

    2024 arXiv

  16. [24]

    Using simulation and domain adaptation to improve efficiency of deep robotic grasping

    Konstantinos Bousmalis, Alex Irpan, Paul Wohlhart, Yunfei Bai, Matthew Kelcey, Mrinal Kalakrishnan, Laura Downs, Julian Ibarz, Peter Pastor, Kurt Konolige, et al. Using simulation and domain adaptation to improve efficiency of deep robotic grasping. In 2018 IEEE inter- nationa...

    2018

  17. [25]

    Dexmimicgen: Automated data generation for bimanual dexterous manipulation via imitation learning

    Zhenyu Jiang, Yuqi Xie, Kevin Lin, Zhenjia Xu, Weikang Wan, Ajay Mandlekar, Linxi Fan, and Yuke Zhu. Dexmimicgen: Automated data generation for bimanual dexterous manipulation via imitation learning. arXiv preprint arXiv:2410.24185, 2024

    arXiv

  18. [26]

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

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choromanski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, et al. Rt-2: Vision-language-action models transfer web knowledge to robotic control. arXiv preprint arXiv:2307.15818 , 2023

    2023 arXiv

  19. [27]

    Dexcap: Scalable and portable mocap data collection system for dexterous manipulation

    Chen Wang, Haochen Shi, Weizhuo Wang, Ruohan Zhang, Li Fei-Fei, and C Karen Liu. Dexcap: Scalable and portable mocap data collection system for dexterous manipulation. arXiv preprint arXiv:2403.07788 , 2024

    2024 arXiv

  20. [28]

    Towards generalizable zero-shot manipulation via translating human interaction plans

    Homanga Bharadhwaj, Abhinav Gupta, Vikash Kumar, and Shubham Tulsiani. Towards generalizable zero-shot manipulation via translating human interaction plans. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 6904–6911. IEEE, 2024

    2024

  21. [29]

    Hrp: Human affordances for robotic pre-training

    Mohan Kumar Srirama, Sudeep Dasari, Shikhar Bahl, and Abhinav Gupta. Hrp: Human affordances for robotic pre-training. arXiv preprint arXiv:2407.18911 , 2024

    2024 arXiv

  22. [30]

    Arcap: Collecting high-quality human demonstrations for robot learning with augmented reality feedback

    Sirui Chen, Chen Wang, Kaden Nguyen, Li Fei-Fei, and C Karen Liu. Arcap: Collecting high-quality human demonstrations for robot learning with augmented reality feedback. arXiv preprint arXiv:2410.08464 , 2024

    2024 arXiv

  23. [31]

    Egomimic: Scaling imitation learning via egocentric video

    Simar Kareer, Dhruv Patel, Ryan Punamiya, Pranay Mathur, Shuo Cheng, Chen Wang, Judy Hoffman, and Danfei Xu. Egomimic: Scaling imitation learning via egocentric video. arXiv preprint arXiv:2410.24221 , 2024

    2024 arXiv

  24. [32]

    Humanoid policy˜ human policy

    Ri-Zhao Qiu, Shiqi Yang, Xuxin Cheng, Chaitanya Chawla, Jialong Li, Tairan He, Ge Yan, David J Yoon, Ryan Hoque, Lars Paulsen, et al. Humanoid policy˜ human policy. arXiv preprint arXiv:2503.13441 , 2025

    2025

  25. [33]

    Bimanual dexterity for complex tasks

    Kenneth Shaw, Yulong Li, Jiahui Yang, Mohan Kumar Srirama, Ray Liu, Haoyu Xiong, Russell Mendonca, and Deepak Pathak. Bimanual dexterity for complex tasks. In 8th Annual Conference on Robot Learning , 2024

    2024

  26. [34]

    Dexwild: Dexterous human interactions for in-the-wild robot policies

    Tony Tao, Mohan Kumar Srirama, Jason Jingzhou Liu, Kenneth Shaw, and Deepak Pathak. Dexwild: Dexterous human interactions for in-the-wild robot policies. arXiv preprint arXiv:2505.07813, 2025

    2025 arXiv

  27. [35]

    Universal manipulation interface: In-the- wild robot teaching without in-the-wild robots

    Cheng Chi, Zhenjia Xu, Chuer Pan, Eric Cousineau, Benjamin Burchfiel, Siyuan Feng, Russ Tedrake, and Shuran Song. Universal manipulation interface: In-the- wild robot teaching without in-the-wild robots. arXiv preprint arXiv:2402.10329, 2024

    2024 arXiv

  28. [36]

    R3m: A universal visual representation for robot manipulation

    Suraj Nair, Aravind Rajeswaran, Vikash Kumar, Chelsea Finn, and Abhinav Gupta. R3m: A universal visual representation for robot manipulation. arXiv preprint arXiv:2203.12601, 2022

    2022 arXiv

  29. [37]

    Dribblebot: Dynamic legged manipulation in the wild

    Yandong Ji, Gabriel B Margolis, and Pulkit Agrawal. Dribblebot: Dynamic legged manipulation in the wild. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 5155–5162. IEEE, 2023

    2023

  30. [38]

    Circus anymal: A quadruped learning dexterous manipulation with its limbs

    Fan Shi, Timon Homberger, Joonho Lee, Takahiro Miki, Moju Zhao, Farbod Farshidian, Kei Okada, Masayuki Inaba, and Marco Hutter. Circus anymal: A quadruped learning dexterous manipulation with its limbs. In 2021 IEEE International Conference on Robotics and Automation (ICRA), p...

    2021

  31. [39]

    Legs as manipulator: Pushing quadrupedal agility beyond lo- comotion

    Xuxin Cheng, Ashish Kumar, and Deepak Pathak. Legs as manipulator: Pushing quadrupedal agility beyond lo- comotion. In 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages 5106–5112. IEEE, 2023

    2023

  32. [40]

    Learning to open and traverse doors with a legged manipulator

    Mike Zhang, Yuntao Ma, Takahiro Miki, and Marco Hutter. Learning to open and traverse doors with a legged manipulator. arXiv preprint arXiv:2409.04882 , 2024

    2024 arXiv

  33. [41]

    Learning visual quadrupedal loco-manipulation from demonstrations

    Zhengmao He, Kun Lei, Yanjie Ze, Koushil Sreenath, Zhongyu Li, and Huazhe Xu. Learning visual quadrupedal loco-manipulation from demonstrations. arXiv preprint arXiv:2403.20328 , 2024

    2024 arXiv

  34. [42]

    Bayrntune: Adaptive bayesian domain randomization via strategic fine-tuning

    Tianle Huang, Nitish Sontakke, K Niranjan Kumar, Ir- fan Essa, Stefanos Nikolaidis, Dennis W Hong, and Sehoon Ha. Bayrntune: Adaptive bayesian domain randomization via strategic fine-tuning. arXiv preprint arXiv:2310.10606, 2023

    arXiv

  35. [43]

    Learning whole-body manipulation for quadrupedal robot

    Seunghun Jeon, Moonkyu Jung, Suyoung Choi, Beomjoon Kim, and Jemin Hwangbo. Learning whole-body manipulation for quadrupedal robot. IEEE Robotics and Automation Letters , 9(1):699–706, 2023

    2023

  36. [44]

    Perceptive pedipulation with local obstacle avoid- ance

    Jonas Stolle, Philip Arm, Mayank Mittal, and Marco Hutter. Perceptive pedipulation with local obstacle avoid- ance. arXiv preprint arXiv:2409.07195 , 2024

    2024 arXiv

  37. [45]

    Learning general- izable feature fields for mobile manipulation

    Ri-Zhao Qiu, Yafei Hu, Ge Yang, Yuchen Song, Yang Fu, Jianglong Ye, Jiteng Mu, Ruihan Yang, Nikolay Atanasov, Sebastian Scherer, et al. Learning general- izable feature fields for mobile manipulation. arXiv preprint arXiv:2403.07563, 2024

    2024 arXiv

  38. [46]

    Cascaded compositional residual learning for complex interactive behaviors

    K Niranjan Kumar, Irfan Essa, and Sehoon Ha. Cascaded compositional residual learning for complex interactive behaviors. IEEE Robotics and Automation Letters , 8(8): 4601–4608, 2023

    2023

  39. [47]

    Learning multi-agent loco-manipulation for long-horizon quadrupedal pushing

    Yuming Feng, Chuye Hong, Yaru Niu, Shiqi Liu, Yux- iang Yang, Wenhao Yu, Tingnan Zhang, Jie Tan, and Ding Zhao. Learning multi-agent loco-manipulation for long-horizon quadrupedal pushing. arXiv preprint arXiv:2411.07104, 2024

    2024 arXiv

  40. [48]

    Mqe: Unleashing the power of interaction with multi-agent quadruped environment

    Ziyan Xiong, Bo Chen, Shiyu Huang, Wei-Wei Tu, Zhaofeng He, and Yang Gao. Mqe: Unleashing the power of interaction with multi-agent quadruped environment. arXiv preprint arXiv:2403.16015 , 2024

    2024 arXiv

  41. [49]

    Solving multi-entity robotic problems using permutation invariant neural networks

    Tianxu An, Joonho Lee, Marko Bjelonic, Flavio De Vin- centi, and Marco Hutter. Solving multi-entity robotic problems using permutation invariant neural networks. arXiv preprint arXiv:2402.18345 , 2024

    2024 arXiv

  42. [50]

    Multi-agent manipulation via lo- comotion using hierarchical sim2real

    Ofir Nachum, Michael Ahn, Hugo Ponte, Shixiang Gu, and Vikash Kumar. Multi-agent manipulation via lo- comotion using hierarchical sim2real. arXiv preprint arXiv:1908.05224, 2019

    1908 arXiv

  43. [51]

    Re- inforcement learning for collaborative quadrupedal ma- nipulation of a payload over challenging terrain

    Yandong Ji, Bike Zhang, and Koushil Sreenath. Re- inforcement learning for collaborative quadrupedal ma- nipulation of a payload over challenging terrain. In 2021 IEEE 17th International Conference on Automation Science and Engineering (CASE) , pages 899–904. IEEE, 2021

    2021

  44. [52]

    Roboduet: A framework affording mobile- manipulation and cross-embodiment

    Guoping Pan, Qingwei Ben, Zhecheng Yuan, Guangqi Jiang, Yandong Ji, Jiangmiao Pang, Houde Liu, and Huazhe Xu. Roboduet: A framework affording mobile- manipulation and cross-embodiment. arXiv preprint arXiv:2403.17367, 2024

    2024 arXiv

  45. [53]

    Visual whole-body control for legged loco-manipulation

    Minghuan Liu, Zixuan Chen, Xuxin Cheng, Yandong Ji, Ruihan Yang, and Xiaolong Wang. Visual whole-body control for legged loco-manipulation. arXiv preprint arXiv:2402.16796, 2024

    2024 arXiv

  46. [54]

    Gamma: Graspability-aware mobile manipula- tion policy learning based on online grasping pose fusion

    Jiazhao Zhang, Nandiraju Gireesh, Jilong Wang, Xi- aomeng Fang, Chaoyi Xu, Weiguang Chen, Liu Dai, and He Wang. Gamma: Graspability-aware mobile manipula- tion policy learning based on online grasping pose fusion. In 2024 IEEE International Conference on Robotics and Automatio...

    2024

  47. [55]

    Pedipulate: Enabling manipulation skills using a quadruped robot’s leg

    Philip Arm, Mayank Mittal, Hendrik Kolvenbach, and Marco Hutter. Pedipulate: Enabling manipulation skills using a quadruped robot’s leg. In 41st IEEE Conference on Robotics and Automation (ICRA 2024) , 2024

    2024

  48. [56]

    A survey of robot learning from demonstration

    Brenna D Argall, Sonia Chernova, Manuela Veloso, and Brett Browning. A survey of robot learning from demonstration. Robotics and autonomous systems , 57 (5):469–483, 2009

    2009

  49. [57]

    Imitation learning: A survey of learning methods

    Ahmed Hussein, Mohamed Medhat Gaber, Eyad Elyan, and Chrisina Jayne. Imitation learning: A survey of learning methods. ACM Computing Surveys (CSUR) , 50 (2):1–35, 2017

    2017

  50. [58]

    Recent advances in robot learning from demonstration

    Harish Ravichandar, Athanasios S Polydoros, Sonia Chernova, and Aude Billard. Recent advances in robot learning from demonstration. Annual review of control, robotics, and autonomous systems , 3(1):297–330, 2020

    2020

  51. [59]

    Strap: Robot sub-trajectory retrieval for augmented policy learning

    Marius Memmel, Jacob Berg, Bingqing Chen, Abhishek Gupta, and Jonathan Francis. Strap: Robot sub-trajectory retrieval for augmented policy learning. In The Thirteenth International Conference on Learning Representations , 2025

    2025

  52. [60]

    Droid: A large-scale in-thewild robot manipulation dataset

    Alexander Khazatsky, Karl Pertsch, Suraj Nair, Ash- win Balakrishna, Sudeep Dasari, Siddharth Karam- cheti, Soroush Nasiriany, Mohan Kumar Srirama, Lawrence Yunliang Chen, Kirsty Ellis, et al. Droid: A large-scale in-thewild robot manipulation dataset. arXiv preprint arXiv:240...

    2024 arXiv

  53. [61]

    Rt-1: Robotics transformer for real-world control at scale

    Anthony Brohan, Noah Brown, Justice Carbajal, Yev- gen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Jasmine Hsu, et al. Rt-1: Robotics transformer for real-world control at scale. arXiv preprint arXiv:2212.06817, 2022

    2022 arXiv

  54. [62]

    On bringing robots home

    Nur Muhammad Mahi Shafiullah, Anant Rai, Haritheja Etukuru, Yiqian Liu, Ishan Misra, Soumith Chintala, and Lerrel Pinto. On bringing robots home. arXiv preprint arXiv:2311.16098, 2023

    2023 arXiv

  55. [63]

    Spawnnet: Learning general- izable visuomotor skills from pre-trained network

    Xingyu Lin, John So, Sashwat Mahalingam, Fangchen Liu, and Pieter Abbeel. Spawnnet: Learning general- izable visuomotor skills from pre-trained network. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 4781–4787. IEEE, 2024

    2024

  56. [64]

    Robot learning on the job: Human- in-the-loop autonomy and learning during deployment

    Huihan Liu, Soroush Nasiriany, Lance Zhang, Zhiyao Bao, and Yuke Zhu. Robot learning on the job: Human- in-the-loop autonomy and learning during deployment. The International Journal of Robotics Research , page 02783649241273901, 2022

    2022

  57. [65]

    Learning human- to-humanoid real-time whole-body teleoperation

    Tairan He, Zhengyi Luo, Wenli Xiao, Chong Zhang, Kris Kitani, Changliu Liu, and Guanya Shi. Learning human- to-humanoid real-time whole-body teleoperation. arXiv preprint arXiv:2403.04436, 2024

    2024 arXiv

  58. [66]

    Real-time whole-body imitation by humanoid robots and task-oriented teleoperation using an analytical mapping method and quantitative evaluation

    Zhijun Zhang, Yaru Niu, Ziyi Yan, and Shuyang Lin. Real-time whole-body imitation by humanoid robots and task-oriented teleoperation using an analytical mapping method and quantitative evaluation. Applied Sciences, 8 (10):2005, 2018

    2005

  59. [67]

    Support changes during online human motion imitation by a humanoid robot using task specification

    Louise Penna Poubel, Sophie Sakka, Denis ´Cehaji´c, and Denis Creusot. Support changes during online human motion imitation by a humanoid robot using task specification. In 2014 IEEE International Conference on Robotics and Automation (ICRA) , pages 1782–1787. IEEE, 2014

    2014

  60. [68]

    Learning to look around: Enhancing teleoperation and learning with a human-like actuated neck

    Bipasha Sen, Michelle Wang, Nandini Thakur, Aditya Agarwal, and Pulkit Agrawal. Learning to look around: Enhancing teleoperation and learning with a human-like actuated neck. arXiv preprint arXiv:2411.00704 , 2024

    2024 arXiv

  61. [69]

    Mobile-television: Predictive motion priors for humanoid whole-body control

    Chenhao Lu, Xuxin Cheng, Jialong Li, Shiqi Yang, Mazeyu Ji, Chengjing Yuan, Ge Yang, Sha Yi, and Xiaolong Wang. Mobile-television: Predictive motion priors for humanoid whole-body control. arXiv preprint arXiv:2412.07773, 2024

    2024 arXiv

  62. [70]

    Bunny-visionpro: Real-time bimanual dexterous teleoperation for imitation learning

    Runyu Ding, Yuzhe Qin, Jiyue Zhu, Chengzhe Jia, Shiqi Yang, Ruihan Yang, Xiaojuan Qi, and Xiaolong Wang. Bunny-visionpro: Real-time bimanual dexterous teleoperation for imitation learning. arXiv preprint arXiv:2407.03162, 2024

    2024 arXiv

  63. [71]

    Airexo: Low-cost exoskeletons for learning whole- arm manipulation in the wild

    Hongjie Fang, Hao-Shu Fang, Yiming Wang, Jieji Ren, Jingjing Chen, Ruo Zhang, Weiming Wang, and Cewu Lu. Airexo: Low-cost exoskeletons for learning whole- arm manipulation in the wild. In 2024 IEEE International Conference on Robotics and Automation (ICRA) , pages 15031–15038....

    2024

  64. [72]

    R+ x: Retrieval and exe- cution from everyday human videos

    Georgios Papagiannis, Norman Di Palo, Pietro Vi- tiello, and Edward Johns. R+ x: Retrieval and exe- cution from everyday human videos. arXiv preprint arXiv:2407.12957, 2024

    2024 arXiv

  65. [73]

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

    Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Foster, Grace Lam, Pannag Sanketi, et al. Openvla: An open-source vision-language-action model. arXiv preprint arXiv:2406.09246, 2024

    2024 arXiv

  66. [74]

    Octo: An open-source generalist robot policy

    Octo Model Team, Dibya Ghosh, Homer Walke, Karl Pertsch, Kevin Black, Oier Mees, Sudeep Dasari, Joey Hejna, Tobias Kreiman, Charles Xu, et al. Octo: An open-source generalist robot policy. arXiv preprint arXiv:2405.12213, 2024

    2024 arXiv

  67. [75]

    Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collaboration 0

    Abby O’Neill, Abdul Rehman, Abhiram Maddukuri, Ab- hishek Gupta, Abhishek Padalkar, Abraham Lee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, Ajinkya Jain, et al. Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collaboration 0. In 2024 IEEE Intern...

    2024

  68. [76]

    Mirage: Cross-embodiment zero-shot policy transfer with cross-painting

    Lawrence Yunliang Chen, Kush Hari, Karthik Dharmara- jan, Chenfeng Xu, Quan Vuong, and Ken Goldberg. Mirage: Cross-embodiment zero-shot policy transfer with cross-painting. arXiv preprint arXiv:2402.19249 , 2024

    2024 arXiv

  69. [77]

    Scaling cross-embodied learning: One policy for manipulation, navigation, locomotion and aviation

    Ria Doshi, Homer Walke, Oier Mees, Sudeep Dasari, and Sergey Levine. Scaling cross-embodied learning: One policy for manipulation, navigation, locomotion and aviation. arXiv preprint arXiv:2408.11812 , 2024

    2024 arXiv

  70. [78]

    Scaling proprioceptive-visual learning with heterogeneous pre-trained transformers

    Lirui Wang, Xinlei Chen, Jialiang Zhao, and Kaim- ing He. Scaling proprioceptive-visual learning with heterogeneous pre-trained transformers. arXiv preprint arXiv:2409.20537, 2024

    2024 arXiv

  71. [79]

    π0: A vision- language-action flow model for general robot control

    Kevin Black, Noah Brown, Danny Driess, Adnan Es- mail, 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. arXiv preprint arXiv:2410.24164 , 2024

    arXiv

  72. [80]

    Gemini robotics: Bringing ai into the physical world

    Gemini Robotics Team, Saminda Abeyruwan, Joshua Ainslie, Jean-Baptiste Alayrac, Montserrat Gonzalez Arenas, Travis Armstrong, Ashwin Balakrishna, Robert Baruch, Maria Bauza, Michiel Blokzijl, et al. Gemini robotics: Bringing ai into the physical world. arXiv preprint arXiv:250...

    2025 arXiv

  73. [81]

    π0.5: a vision-language-action model with open-world generalization

    Physical Intelligence, Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Es- mail, Michael Equi, Chelsea Finn, Niccolo Fusai, et al. π0.5: a vision-language-action model with open-world generalization. arXiv preprint arXiv:2504.16054 , 2025

    2025 arXiv

  74. [82]

    Gr00t n1: An open foundation model for generalist humanoid robots

    Johan Bjorck, Fernando Casta ˜neda, Nikita Cherniadev, Xingye Da, Runyu Ding, Linxi Fan, Yu Fang, Dieter Fox, Fengyuan Hu, Spencer Huang, et al. Gr00t n1: An open foundation model for generalist humanoid robots. arXiv preprint arXiv:2503.14734, 2025

    2025 arXiv

  75. [83]

    The pinocchio c++ library: A fast and flexible implementation of rigid body dynamics algorithms and their analytical derivatives

    Justin Carpentier, Guilhem Saurel, Gabriele Buondonno, Joseph Mirabel, Florent Lamiraux, Olivier Stasse, and Nicolas Mansard. The pinocchio c++ library: A fast and flexible implementation of rigid body dynamics algorithms and their analytical derivatives. In 2019 IEEE/SICE Int...

    2019

  76. [84]

    The hyper- parameters used for training are summarized in Table VII

    is based on the released codebase, with only minor TABLE VIII: HPT hyperparameters Hyperparameters Value optimizer AdamW learning rate 5e-5 (finetuning/from scratch) 1e-4 (pretraining) scheduler constant weight decay 1e-4 trunk #Transformer blocks 16 hidden dim 128 feedforward...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.