pith. sign in

arxiv: 2605.15536 · v1 · pith:LW2MF4YFnew · submitted 2026-05-15 · 💻 cs.RO · cs.AI· cs.CV

SkiP: When to Skip and When to Refine for Efficient Robot Manipulation

Mingtong Dai , Guanqi Peng , Yongjie Bai , Feng Yan , Chunjie Chen , Lingbo Liu , Liang Lin , Xinyu Wu This is my paper

Pith reviewed 2026-05-19 14:43 UTC · model grok-4.3

classification 💻 cs.RO cs.AIcs.CV
keywords imitation learningrobot manipulationaction skippingefficient controlbehavior cloningmotion analysisskip policy
0
0 comments X p. Extension
pith:LW2MF4YF Add to your LaTeX paper What is a Pith Number? 
\usepackage{pith}
\pithnumber{LW2MF4YF}

Prints a linked pith:LW2MF4YF badge after your title and writes the identifier into PDF metadata. Compiles on arXiv with no extra files. Learn more

The pith

SkiP lets a single robot policy learn to skip low-information steps by relabeling actions to the next key segment.

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

Imitation learning policies typically predict actions at every step, even during smooth free-space motions that carry little information. This paper shows that by relabeling the training targets in those segments to the action at the start of the next key segment, the policy can learn to leap over them in one decision. Key segments around contacts and grasps still receive dense prediction. An automatic method called Motion Spectrum Keying partitions the demonstrations into these segments without manual labels. The result is a flat network that executes fewer steps while matching success rates on dozens of manipulation tasks.

Core claim

The Skip Policy (SkiP) is formed by an action relabeling mechanism where, for each timestep in a skip segment, the behavior cloning target is replaced with the action at the entrance of the next key segment. Combined with Motion Spectrum Keying to detect key and skip segments from action signals, this allows the policy to dynamically skip redundant steps and refine at critical points in a single unified network without any hierarchical structure or learned planner.

What carries the argument

Action relabeling mechanism that points skip-segment targets to the next key action, paired with Motion Spectrum Keying for automatic segment detection.

If this is right

  • Executed steps drop by 15 to 40 percent across 72 simulated tasks and three real-robot tasks.
  • Success rates match or exceed those of standard policies on various backbones.
  • The approach requires no separate skip planner or hierarchical policy structure.
  • Key and skip segments are identified automatically from motion complexity in the demonstrations.

Where Pith is reading between the lines

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

  • Similar relabeling could reduce computation in other long-horizon imitation learning domains such as navigation or assembly.
  • Integrating SkiP with visual observations might further highlight how key segments align with visual features like object contacts.

Load-bearing premise

Replacing the behavior cloning target in skip segments with the action from the next key segment still yields a policy that executes safely in closed-loop control without extra safety or recovery mechanisms.

What would settle it

Running the learned policy on a task where skipping causes the robot to miss a grasp or collide during the leap, while a non-skipping policy succeeds, would falsify the claim.

Figures

Figures reproduced from arXiv: 2605.15536 by Chunjie Chen, Feng Yan, Guanqi Peng, Liang Lin, Lingbo Liu, Mingtong Dai, Xinyu Wu, Yongjie Bai.

Figure 1
Figure 1. Figure 1: “Pick up 1 cup from the mug tree and place it on the table” analyzed by SkiP. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of SkiP. We partition each demonstration into high-information [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a) Illustration of SkiP’s relabeling scheme: in skip segments, the training target jumps to [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Per-task success rates on RLBench-50 (tasks sorted by best SR). SkiP improves over [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ablation on quantile threshold q (RLBench-60, 3 eval repeats, shaded bands = ±1 std). SR peaks at q=0.75 and drops for q ≥ 0.80; Stepssucc decreases monotonically. Quantile threshold q. The threshold q controls how conservatively key segments are labeled: larger q marks fewer timesteps as refine-worthy [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Action displacement distribution per policy call across 10 RLBench tasks. SkiP shows a [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: shows the real-robot setup and rollout examples for the three tabletop tasks. These examples match the tasks used in [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
read the original abstract

Previous imitation learning policies predict future actions at every control step, whether in smooth motion phases or precise, contact-rich operation phases. This uniform treatment is wasteful: most steps in a manipulation trajectory traverse free space and carry little task-relevant information, while a small fraction of \emph{key} steps around contacts, grasps, and alignment demand dense, high-resolution prediction. We propose a novel \emph{action relabeling} mechanism: at each timestep in a skip segment, we replace the behavior cloning target with the action at the entrance of the next key segment, enabling the policy to leap over redundant steps in a single decision. The resulting \textbf{Skip Policy (SkiP)} dynamically leaps over skip segments and intensively refines actions in key segments, within a single unified network requiring no learned skip planner or hierarchical structure. To automatically partition demonstrations into key and skip segments without manual annotation, we introduce \emph{Motion Spectrum Keying} (MSK), a fast, task-agnostic procedure that detects local motion complexity from action signals. Extensive experiments across 72 simulated manipulation tasks and three real-robot tasks show that SkiP reduces executed steps by $15$--$40\%$ while matching or improving success rates across various policy backbones. Project page: \texttt{https://pgq18.github.io/SkiP-page/}.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 3 minor

Summary. The manuscript introduces SkiP, an imitation learning method for robot manipulation that automatically segments demonstration trajectories into key and skip segments via Motion Spectrum Keying (MSK) and applies action relabeling during behavior cloning. In skip segments, the policy is trained to output the action at the entrance of the next key segment, enabling a single unified network to leap over redundant steps at execution time without a separate skip planner or hierarchy. Experiments report 15-40% reduction in executed steps with matched or improved success rates across 72 simulated tasks and three real-robot tasks.

Significance. If the empirical claims hold under closed-loop execution, SkiP provides a lightweight way to improve efficiency in behavior-cloning policies by concentrating action prediction on contact-rich phases. The scale of the evaluation (72 simulated tasks plus real-robot validation) and the absence of extra learned components are concrete strengths that would make the result practically relevant for resource-constrained manipulation.

major comments (1)
  1. [§3.2] §3.2 (Action Relabeling): the training procedure replaces the BC target at every timestep inside a skip segment with the action a_k at the entrance of the next key segment. The manuscript provides no analysis or additional experiments demonstrating that the resulting policy remains stable and collision-free when real states deviate from the demonstration distribution during closed-loop execution of these temporally distant actions. This assumption is load-bearing for the central claim that a single unified network suffices without recovery behaviors or safety filters.
minor comments (3)
  1. [Figure 3] Figure 3 and §4.2: the caption and text do not clarify whether the reported step reductions are measured in open-loop replay or closed-loop execution; adding this distinction would improve reproducibility.
  2. [§4.1] §4.1: the description of MSK threshold selection is brief; a short sensitivity plot or explicit default values would help readers replicate the segmentation on new tasks.
  3. [Table 2] Table 2: several rows report success-rate improvements without accompanying standard deviations or number of trials; adding these statistics would strengthen the cross-task claims.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the practical relevance of SkiP's lightweight approach. We address the single major comment below with clarifications drawn from the existing evaluation and indicate where we will strengthen the manuscript.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Action Relabeling): the training procedure replaces the BC target at every timestep inside a skip segment with the action a_k at the entrance of the next key segment. The manuscript provides no analysis or additional experiments demonstrating that the resulting policy remains stable and collision-free when real states deviate from the demonstration distribution during closed-loop execution of these temporally distant actions. This assumption is load-bearing for the central claim that a single unified network suffices without recovery behaviors or safety filters.

    Authors: We thank the referee for highlighting this important robustness consideration. The 72 simulated tasks and three real-robot tasks were all evaluated under closed-loop execution, where the policy receives observed states that can deviate from the demonstration distribution due to sensor noise, dynamics mismatch, and compounding errors. In these experiments SkiP maintains or improves success rates while cutting executed steps by 15-40 percent, providing direct empirical evidence that the relabeled targets do not produce unstable or colliding behavior in practice. Motion Spectrum Keying further restricts skip segments to low-complexity, smooth free-space motions, reducing the risk associated with temporally distant actions. We nevertheless agree that an explicit discussion of distribution-shift robustness would strengthen the paper. In the revision we will add a dedicated paragraph in Section 5 that (i) summarizes failure modes observed on the real robot and (ii) provides qualitative trajectory overlays showing that state deviations within skip segments did not result in collisions or unsafe motions. This addition grounds the central claim in the existing large-scale results while directly addressing the referee's concern. revision: partial

Circularity Check

0 steps flagged

No significant circularity: empirical method with independent experimental validation

full rationale

The SkiP paper presents an empirical approach consisting of a Motion Spectrum Keying heuristic to segment demonstrations and an action relabeling step that modifies behavior-cloning targets for skip segments. These are training modifications whose effects on execution efficiency and success rate are measured directly in 72 simulated tasks plus real-robot experiments. No equations, uniqueness theorems, or self-citations are invoked to derive the performance gains; the reported 15-40% step reduction is an observed outcome rather than a quantity forced by construction from fitted parameters or prior author results. The central claim therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method rests on standard imitation learning assumptions and introduces two new procedural components (relabeling rule and MSK) without additional free parameters or invented physical entities.

axioms (1)
  • domain assumption Demonstration trajectories contain identifiable segments of low motion complexity that can be safely skipped without affecting task success.
    Invoked in the description of MSK and the relabeling mechanism.

pith-pipeline@v0.9.0 · 5798 in / 1186 out tokens · 63650 ms · 2026-05-19T14:43:37.646871+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages · 10 internal anchors

  1. [1]

    Sail: Faster-than-demonstration execution of imitation learning policies, 2025

    Nadun Ranawaka Arachchige, Zhenyang Chen, Wonsuhk Jung, Woo Chul Shin, Rohan Bansal, Pierre Barroso, Yu Hang He, Yingyang Celine Lin, Benjamin Joffe, Shreyas Kousik, and Danfei Xu. Sail: Faster-than-demonstration execution of imitation learning policies, 2025. URL https://arxiv.org/abs/2506.11948

    work page arXiv 2025

  2. [2]

    Learning to see and act: Task-aware virtual view exploration for robotic manipulation, 2025

    Yongjie Bai, Zhouxia Wang, Yang Liu, Kaijun Luo, Yifan Wen, Mingtong Dai, Weixing Chen, Ziliang Chen, Lingbo Liu, Guanbin Li, and Liang Lin. Learning to see and act: Task-aware virtual view exploration for robotic manipulation, 2025. URL https://arxiv.org/abs/ 2508.05186

    work page arXiv 2025

  3. [3]

    $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control

    Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Lucy Xiaoyang Shi, James Tanner, Quan Vuong, Anna Walling, Haohuan Wang, and Ury Zhilinsky. π0: A visi...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  4. [4]

    RT-1: Robotics Transformer for Real-World Control at Scale

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alexander Herzog, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Tomas Jackson, Sally Jesmonth, Nikhil Joshi, Ryan Julian, Dmitry Kalashnikov, Yuheng Kuang, Isabel Leal, Kuang-Huei Lee, Sergey Levine, Yao Lu, Utsav Mall...

    work page doi:10.15607/rss.2023.xix.025 2023

  5. [5]

    Diffusion policy: Visuomotor policy learning via action diffusion,

    Cheng Chi, Zhenjia Xu, Siyuan Feng, Eric Cousineau, Yilun Du, Benjamin Burchfiel, Russ Tedrake, and Shuran Song. Diffusion policy: Visuomotor policy learning via action diffusion,

    work page

  6. [6]

    URLhttps://arxiv.org/abs/2303.04137

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    Act3d: 3d feature field transformers for multi-task robotic manipulation

    Theophile Gervet, Zhou Xian, Nikolaos Gkanatsios, and Katerina Fragkiadaki. Act3d: 3d feature field transformers for multi-task robotic manipulation. In Jie Tan, Marc Toussaint, and Kourosh Darvish, editors,Proceedings of The 7th Conference on Robot Learning, volume 229 ofProceedings of Machine Learning Research, pages 3949–3965. PMLR, 06–09 Nov 2023. URL...

    work page 2023

  8. [8]

    Gemini 2.5: Our most intelligent AI model, 2025

    Google DeepMind. Gemini 2.5: Our most intelligent AI model, 2025. URL https:// deepmind.google/technologies/gemini/. Technical report

    work page 2025

  9. [9]

    Rvt: Robotic view transformer for 3d object manipulation

    Ankit Goyal, Jie Xu, Yijie Guo, Valts Blukis, Yu-Wei Chao, and Dieter Fox. Rvt: Robotic view transformer for 3d object manipulation. In Jie Tan, Marc Toussaint, and Kourosh Darvish, editors,Proceedings of The 7th Conference on Robot Learning, volume 229 ofProceedings of Machine Learning Research, pages 694–710. PMLR, 06–09 Nov 2023. URL https: //proceedin...

    work page 2023

  10. [10]

    Instruction-driven history-aware policies for robotic manipulations

    Pierre-Louis Guhur, Shizhe Chen, Ricardo Garcia Pinel, Makarand Tapaswi, Ivan Laptev, and Cordelia Schmid. Instruction-driven history-aware policies for robotic manipulations. In Karen Liu, Dana Kulic, and Jeff Ichnowski, editors,Proceedings of The 6th Conference on Robot Learning, volume 205 ofProceedings of Machine Learning Research, pages 175–187. PMLR...

    work page 2023

  11. [11]

    DaDu-Corki: Algorithm-architecture co-design for embodied AI-powered robotic manipulation, 2025

    Yiyang Huang, Yuhui Hao, Bo Yu, Feng Yan, Yuxin Yang, Feng Min, Yinhe Han, Lin Ma, Shaoshan Liu, Qiang Liu, and Yiming Gan. DaDu-Corki: Algorithm-architecture co-design for embodied AI-powered robotic manipulation, 2025. URL https://arxiv.org/abs/2407. 04292. 10

    work page 2025

  12. [12]

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

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [13]

    Coarse-to-fine q-attention with learned path ranking, 2022

    Stephen James and Pieter Abbeel. Coarse-to-fine q-attention with learned path ranking, 2022. URLhttps://arxiv.org/abs/2204.01571

    work page arXiv 2022

  14. [14]

    Stephen James, Zicong Ma, David Rovick Arrojo, and Andrew J. Davison. Rlbench: The robot learning benchmark & learning environment.IEEE Robotics and Automation Letters, 5(2): 3019–3026, apr 2020. doi: 10.1109/LRA.2020.2974707. URL https://doi.org/10.1109/ LRA.2020.2974707

    work page doi:10.1109/lra.2020.2974707 2020

  15. [15]

    3D Diffuser Actor: Policy Diffusion with 3D Scene Representations

    Tsung-Wei Ke, Nikolaos Gkanatsios, and Katerina Fragkiadaki. 3D diffuser actor: Policy diffusion with 3D scene representations, 2024. URL https://arxiv.org/abs/2402.10885

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  16. [16]

    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, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. OpenVLA: An open-source vision-language-action model, 2024. URL https://arxiv. org/abs/...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  17. [17]

    KISA: A unified keyframe identifier and skill annotator for long-horizon robotics demonstrations

    Longxin Kou, Fei Ni, Yan Zheng, Jinyi Liu, Yifu Yuan, Zibin Dong, and Jianye Hao. KISA: A unified keyframe identifier and skill annotator for long-horizon robotics demonstrations. In Ruslan Salakhutdinov, Zico Kolter, Katherine Heller, Adrian Weller, Nuria Oliver, Jonathan Scarlett, and Felix Berkenkamp, editors,Proceedings of the 41st International Confe...

    work page 2024

  18. [18]

    RoboUni- View: Visual-language model with unified view representation for robotic manipulation, 2024

    Fanfan Liu, Feng Yan, Liming Zheng, Chengjian Feng, Yiyang Huang, and Lin Ma. RoboUni- View: Visual-language model with unified view representation for robotic manipulation, 2024. URLhttps://arxiv.org/abs/2406.18977

    work page arXiv 2024

  19. [19]

    What matters in learning from offline human demonstrations for robot manipulation

    Ajay Mandlekar, Danfei Xu, Josiah Wong, Soroush Nasiriany, Chen Wang, Rohun Kulkarni, Li Fei-Fei, Silvio Savarese, Yuke Zhu, and Roberto Mart´ın-Mart´ın. What matters in learning from offline human demonstrations for robot manipulation. InProceedings of The 5th Confer- ence on Robot Learning, volume 164 ofProceedings of Machine Learning Research, pages 16...

    work page 2022

  20. [20]

    CALVIN: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks.IEEE Robotics and Automation Letters, 7(3):7327–7334, 2022

    Oier Mees, Lukas Hermann, Erick Rosete-Beas, and Wolfram Burgard. CALVIN: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks.IEEE Robotics and Automation Letters, 7(3):7327–7334, 2022. doi: 10.1109/LRA.2022.3180108. URLhttps://doi.org/10.1109/LRA.2022.3180108

    work page doi:10.1109/lra.2022.3180108 2022

  21. [21]

    Robotwin: Dual-arm robot benchmark with generative digital twins, 2024

    Yao Mu, Tianxing Chen, Shijia Peng, Zanxin Chen, Zeyu Gao, Yude Zou, Lunkai Lin, Zhiqiang Xie, and Ping Luo. Robotwin: Dual-arm robot benchmark with generative digital twins, 2024. URLhttps://arxiv.org/abs/2409.02920

    work page arXiv 2024

  22. [22]

    Open X-Embodiment: Robotic Learning Datasets and RT-X Models

    Open X-Embodiment Collaboration et al. Open x-embodiment: Robotic learning datasets and RT-X models, 2023. URLhttps://arxiv.org/abs/2310.08864

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  23. [23]

    FAST: Efficient Action Tokenization for Vision-Language-Action Models

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

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  24. [24]

    Efficient reductions for imitation learning

    Stephane Ross and Drew Bagnell. Efficient reductions for imitation learning. In Yee Whye Teh and Mike Titterington, editors,Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, volume 9 ofProceedings of Machine Learning Research, pages 661–668, Chia Laguna Resort, Sardinia, Italy, 13–15 May 2010. PMLR. URL http...

    work page 2010

  25. [25]

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

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

    work page 2011

  26. [26]

    Behavior transformers: Cloning k modes with one stone, 2022

    Nur Muhammad Mahi Shafiullah, Zichen Jeff Cui, Ariuntuya Altanzaya, and Lerrel Pinto. Behavior transformers: Cloning k modes with one stone, 2022. URL https://arxiv.org/ abs/2206.11251

    work page arXiv 2022

  27. [27]

    CLIPort: What and where pathways for robotic manipulation, 2021

    Mohit Shridhar, Lucas Manuelli, and Dieter Fox. CLIPort: What and where pathways for robotic manipulation, 2021. URLhttps://arxiv.org/abs/2109.12098

    work page arXiv 2021

  28. [28]

    Perceiver-actor: A multi-task transformer for robotic manipulation

    Mohit Shridhar, Lucas Manuelli, and Dieter Fox. Perceiver-actor: A multi-task transformer for robotic manipulation. In Karen Liu, Dana Kulic, and Jeff Ichnowski, editors,Proceedings of The 6th Conference on Robot Learning, volume 205 ofProceedings of Machine Learning Research, pages 785–799. PMLR, 14–18 Dec 2023. URL https://proceedings.mlr.press/v205/ sh...

    work page 2023

  29. [29]

    Freqpolicy: Efficient flow-based visuomotor policy via frequency consistency,

    Yifei Su, Ning Liu, Dong Chen, Zhen Zhao, Kun Wu, Meng Li, Zhiyuan Xu, Zhengping Che, and Jian Tang. Freqpolicy: Efficient flow-based visuomotor policy via frequency consistency,

    work page

  30. [30]

    URLhttps://arxiv.org/abs/2506.08822

    work page arXiv

  31. [31]

    Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning.Artificial Intelligence, 112(1-2): 181–211, 1999

    Richard S Sutton, Doina Precup, and Satinder Singh. Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning.Artificial Intelligence, 112(1-2): 181–211, 1999

    work page 1999

  32. [32]

    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, Jianlan Luo, You Liang Tan, Lawrence Yunliang Chen, Pannag Sanketi, Quan Vuong, Ted Xiao, Dorsa Sadigh, Chelsea Finn, and Sergey Levine. Octo: An open-source generalist robot policy, 2024. URL https://arxiv.org/abs/2405.12213

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  33. [33]

    Keyframe-focused visual imitation learning, 2021

    Chuan Wen, Jierui Lin, Jianing Qian, Yang Gao, and Dinesh Jayaraman. Keyframe-focused visual imitation learning, 2021. URLhttps://arxiv.org/abs/2106.06452

    work page arXiv 2021

  34. [34]

    Chaineddiffuser: Unifying trajectory diffusion and keypose prediction for robotic manipulation

    Zhou Xian, Nikolaos Gkanatsios, Theophile Gervet, Tsung-Wei Ke, and Katerina Fragkiadaki. Chaineddiffuser: Unifying trajectory diffusion and keypose prediction for robotic manipulation. In Jie Tan, Marc Toussaint, and Kourosh Darvish, editors,Proceedings of The 7th Conference on Robot Learning, volume 229 ofProceedings of Machine Learning Research, pages ...

    work page 2023

  35. [35]

    RoboTron-Mani: All-in-one multimodal large model for robotic manipula- tion, 2025

    Feng Yan, Fanfan Liu, Liming Zheng, Yufeng Zhong, Yiyang Huang, Zechao Guan, Chengjian Feng, and Lin Ma. RoboTron-Mani: All-in-one multimodal large model for robotic manipula- tion, 2025. URLhttps://arxiv.org/abs/2412.07215

    work page arXiv 2025

  36. [36]

    Wavelet Policy: Imitation Learning in the Scale Domain with World Prior Memory

    Changchuan Yang, Yuhang Dong, Guanzhong Tian, Haizhou Ge, and Hongrui Zhu. Wavelet policy: Imitation policy learning in the scale domain with wavelet transforms, 2025. URL https://arxiv.org/abs/2504.04991

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  37. [37]

    3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations

    Yanjie Ze, Gu Zhang, Kangning Zhang, Chenyuan Hu, Muhan Wang, and Huazhe Xu. 3d diffusion policy: Generalizable visuomotor policy learning via simple 3d representations, 2024. URLhttps://arxiv.org/abs/2403.03954

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  38. [38]

    Chain-of-action: Trajectory autoregressive modeling for robotic manipulation, 2025

    Wenbo Zhang, Tianrun Hu, Yanyuan Qiao, Hanbo Zhang, Yuchu Qin, Yang Li, Jiajun Liu, Tao Kong, Lingqiao Liu, and Xiao Ma. Chain-of-action: Trajectory autoregressive modeling for robotic manipulation, 2025. URLhttps://arxiv.org/abs/2506.09990

    work page arXiv 2025

  39. [39]

    Autore- gressive action sequence learning for robotic manipulation, 2025

    Xinyu Zhang, Yuhan Liu, Haonan Chang, Liam Schramm, and Abdeslam Boularias. Autore- gressive action sequence learning for robotic manipulation, 2025. URL https://arxiv.org/ abs/2410.03132. 12

    work page arXiv 2025

  40. [40]

    Zhiyuan Zhou, Andy Peng, Qiyang Li, Sergey Levine, and Aviral Kumar

    Tony Z. Zhao, Vikash Kumar, Sergey Levine, and Chelsea Finn. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. InProceedings of Robotics: Science and Systems, Daegu, Republic of Korea, July 2023. doi: 10.15607/RSS.2023.XIX.016

    work page doi:10.15607/rss.2023.xix.016 2023

  41. [41]

    RoboCAS: A benchmark for robotic manipulation in complex object arrangement scenarios, 2024

    Liming Zheng, Feng Yan, Fanfan Liu, Chengjian Feng, Zhuoliang Kang, and Lin Ma. RoboCAS: A benchmark for robotic manipulation in complex object arrangement scenarios, 2024. URL https://arxiv.org/abs/2407.06951

    work page arXiv 2024

  42. [42]

    Freqpolicy: Frequency autoregressive visuomotor policy with continuous tokens, 2025

    Yiming Zhong, Yumeng Liu, Chuyang Xiao, Zemin Yang, Youzhuo Wang, Yufei Zhu, Ye Shi, Yujing Sun, Xinge Zhu, and Yuexin Ma. Freqpolicy: Frequency autoregressive visuomotor policy with continuous tokens, 2025. URLhttps://arxiv.org/abs/2506.01583

    work page arXiv 2025

  43. [43]

    Sanketi, Grecia Salazar, Michael S

    Brianna Zitkovich, Tianhe Yu, Sichun Xu, Peng Xu, Ted Xiao, Fei Xia, Jialin Wu, Paul Wohlhart, Stefan Welker, Ayzaan Wahid, Quan Vuong, Vincent Vanhoucke, Huong Tran, Radu Soricut, Anikait Singh, Jaspiar Singh, Pierre Sermanet, Pannag R. Sanketi, Grecia Salazar, Michael S. Ryoo, Krista Reymann, Kanishka Rao, Karl Pertsch, Igor Mordatch, Henryk Michalewski...

    work page 2023