Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

AnchorDP3: 3D Affordance Guided Sparse Diffusion Policy for Robotic Manipulation

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read AnchorDP3 claims a diffusion policy that predicts only sparse, affordance-anchored keyposes achieves a 98.7% average success rate on the RoboTwin benchmark's randomized dual-arm manipulation tasks.

desk verdict A coherent competition-winning architecture whose headline number is plausible but not pinned down in the text; would benefit from a serious referee if the authors add evaluation details. read the letter →

arxiv 2506.19269 v2 pith:7KMITY63 submitted 2025-06-24 cs.RO cs.AI

Ziyan Zhao , Ke Fan , He-Yang Xu , Ning Qiao , Bo Peng , Wenlong Gao , Dongjiang Li , Hui Shen This is my paper
classification cs.ROcs.AI
keywords dual-armmanipulationdiffusionpolicyaffordanceanchoringkeyposepredictionpointcloudperceptionmulti-tasklearningsimulation-to-realsparseactionrepresentation
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

AnchorDP3 sets out to show that a dual-arm manipulation policy can be trained entirely in simulation and still handle extreme scene randomization by learning from sparse, geometrically meaningful action anchors instead of dense trajectories. Its central claim is a 98.7% average success rate across the RoboTwin benchmark tasks under randomized objects, clutter, table height, lighting, and backgrounds, with the top scores in both simulated tracks of the RoboTwin Challenge. The paper argues that three design choices carry the result: simulator-rendered semantic segmentation of task-critical objects, per-task feature encoders feeding a shared diffusion action expert, and keypose prediction anchored to object affordances with full-state supervision. If the result holds, it would suggest that human demonstrations can be replaced by procedurally generated expert data plus an affordance-aware sparse action formulation.

What carries the argument

The load-bearing mechanism is the 'action anchor': a geometric keypose such as pre-grasp or grasp that is defined as an affine function of an object's affordance, for example affordance centroid plus offset. Instead of emitting dense 20-25 Hz trajectories, the diffusion expert—a Conditional U-Net 1D with FiLM conditioning—predicts a list of H=8 future action anchors, and only the first anchor is executed. This compresses the action space to 10-30 anchors per trajectory, and the dataset packs each observation with a list of future anchors, allowing 14.5 times more trajectories within the same compute budget. The 32-dimensional per-anchor state includes both joint angles and end-effector poses, forcing the network to learn geometrically consistent full-state predictions even though only the joint and gripper dimensions are executed.

What would settle it

Run the official RoboTwin evaluation protocol on the original six task categories with the trained policy and report per-task success rates with episode counts; if the average falls materially below 98.7%, or if any single task category drops below the previous best published result, the benchmark claim is falsified.

Watch

Extended reading notes

Core claim

The paper's central discovery is that replacing dense, high-frequency action prediction with sparse affordance-anchored keyposes—pre-grasp, grasp, place, and related transition poses—drastically simplifies the learning problem without sacrificing task performance. The policy predicts an eight-step horizon of 32-dimensional action anchors, each containing joint angles and end-effector poses with rotations encoded by the first two columns of the rotation matrix, using a single shared Conditional U-Net conditioned by one of eight task-specific point-cloud encoders. Simulator-supervised segmentation provides per-point labels for task-critical objects, and the training set is built by tagging atomic expert-policy boundaries as anchors, randomly sampling only two frames between anchors, and injecting DAgger-like recovery observations with 10% probability. On the RoboTwin benchmark, the authors report a 98.7% average success rate and state this was the highest result in the competition's simulation tracks I and II.

Load-bearing premise

The headline 98.7% success rate is reported as a single mean over a modified 8-task split of the benchmark, without per-task counts or variance, so the claim of top competition performance depends on that number matching the official evaluation protocol.

Editorial extensions

If this is right

  • If the 98.7% result holds under the official evaluation protocol, simulation-only training with procedural data becomes a viable alternative to human-demonstration pipelines for dual-arm manipulation.
  • The sparse anchor dataset protocol yields 14.5 times more trajectories for the same compute, so scaling environment diversity becomes substantially cheaper.
  • Predicting only affordance-anchored keyposes reduces the action space from thousands of dense actions to tens of anchors per task, which should ease long-horizon credit assignment and reduce spurious correlations with past motion.
  • The DAgger-style injection of recovery observations during rendering adds robustness against suboptimal states, a property that carries over to deployment.
  • Combined with the RoboTwin real-to-sim pipeline, the authors argue the framework could generate deployable policies from scene and instruction alone, eliminating human demonstrations from the learning loop.

Reading between the lines

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

  • The anchor abstraction is not intrinsically tied to RoboTwin's modular expert: the paper notes that keyposes can be detected in human demonstrations via slowdowns at motion transitions, so the same idea could in principle be applied to unsegmented demonstration video, which the authors did not implement.
  • Because only the first predicted anchor is executed, the policy is effectively a reactive controller with an eight-step lookahead; a testable extension is whether a shorter horizon suffices once affordances are segmented, which would further reduce compute.
  • The design of keeping perception specialized per task while sharing the action expert suggests a general recipe for multi-task visuomotor learning that may transfer beyond this benchmark.
  • Readers verifying the benchmark claim should check whether the 98.7% average was computed on the paper's modified 8-task split rather than the original 6-task competition protocol, since the paper's own taxonomy change makes direct leaderboard comparison non-trivial.
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 / 5 minor

Summary. AnchorDP3 is a diffusion-policy framework for dual-arm robotic manipulation in the RoboTwin simulation benchmark. The paper proposes three main components: simulator-supervised semantic segmentation that labels task-critical objects from rendered depth differences, task-conditioned point-cloud encoders that feed a shared diffusion action expert, and affordance-anchored keypose diffusion that predicts sparse action anchors (pre-grasp, grasp, place, etc.) instead of dense trajectories. The policy also uses an augmented 32-dimensional state/action representation and a DAgger-style data collection protocol. The central claim is a 98.7% average success rate on the RoboTwin benchmark and the highest success rate in both simulation tracks of the RoboTwin Challenge.

Significance. If the reported performance is substantiated, AnchorDP3 would be a strong result in simulated bimanual manipulation under heavy randomization. The paper offers a sensible and clearly described recipe: rendering-derived semantic masks for affordance awareness, per-task encoders to reduce multi-task interference, and sparse keypose prediction to simplify the action space. These ideas are plausible and could be useful to the community. However, the manuscript as submitted is a competition technical report rather than a complete empirical study. The head-line number is presented as a single aggregate with no per-task breakdown, episode counts, error bars, or comparison baselines, and the task taxonomy used for evaluation is ambiguous because Section 3.2 modifies the official 6-task split into 8 categories. I do not view the use of RoboTwin's own expert pipeline for keypose labeling as circular, since it is a legitimate label-construction choice, but the evaluation protocol needs to be clarified and the evidence substantially expanded before the central claim can be verified.

major comments (3)
  1. [Section 3.2 (Language instruction following)] The paper defines 8 distinct manipulation tasks by splitting the official place_object_scale category into place_mouse, place_stapler, and place_bell, while the RoboTwin benchmark specifies 6 task categories. The abstract and Section 4 report a 98.7% average success rate 'in the RoboTwin benchmark' and 'the highest success rate in the RoboTwin Challenge simulation tracks,' but the text never states whether this number was computed on the official 6-task protocol or on the modified 8-task split. If the latter, the aggregate is not directly comparable to the official benchmark or the competition rankings, and the state-of-the-art claim is not verifiable from the paper. This should be resolved by reporting per-task success rates under the official taxonomy, the number of evaluation episodes per task, and, ideally, the official evaluation logs.
  2. [Section 4 / Abstract] The central empirical claim rests on a single aggregate success rate with no per-task breakdown, no episode counts, no variance, and no named baselines. Without this information, the reader cannot assess whether the aggregate is dominated by easy tasks or by an uneven episode distribution, nor can the 'highest success rate' claim be checked against any concrete competitor. Please add a table with per-task success rates, standard errors or confidence intervals, evaluation episode counts, and a clear statement of the exact protocol used to produce the 98.7% figure.
  3. [Sections 3.2-3.4] Several load-bearing design claims are asserted without ablation evidence. For example, Section 3.3 states that full-state supervision 'is beneficial for converging,' and Section 3.4 states that the increased environmental exposure 'contributes to the improvement of the model's accuracy.' Since the paper's contribution is precisely these design choices, the absence of any ablation of the semantic segmentation module, task-conditioned encoders, keypose diffusion, and DAgger collection makes the contribution difficult to evaluate. At minimum, report an ablation of the three main components on a subset of tasks, or explicitly state if the competition constraints prevented such experiments.
minor comments (5)
  1. [Section 2.2] The sentence 'However, there is hard work that focuses on improving the performance of the diffusion policy algorithm based on the keyposes' is unclear and grammatically awkward; please rewrite.
  2. [Section 3.2] The point-cloud projection equation uses notation inconsistently: R_i^{-1} and t_i are not explicitly defined, and the matrix-vector products are not fully specified. Please clarify the coordinate frames and dimensions.
  3. [Figure 2] The overall workflow diagram is dense and the roles of 'argmax', 'Denoise', and the frozen PointNet++ module are ambiguous. Label the inputs and outputs of each module more clearly.
  4. [Section 3.3] The sentence 'The neural networks do not explicitly learn the names and tags of each action anchor' is vague; please specify that the model predicts a fixed-length sequence of continuous 32-dimensional anchors without explicit symbolic labels.
  5. [References] References [5] and [25] are the same paper (Diffusion Policy), and references [6] and [26] are also duplicate (3D Diffusion Policy). Please merge or cite once.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the 98.7% success rate is an empirical benchmark measurement, not a quantity derived from the paper's own parameters or definitions.

full rationale

AnchorDP3 is a systems/technical-report paper whose central claim is an empirically measured success rate on the RoboTwin benchmark. The paper does not offer a mathematical derivation in which a predicted quantity reduces by construction to a fitted parameter or to a self-citation. The keypose action anchors are defined from RoboTwin's modular expert policy boundaries, and the semantic segmentation labels are generated from rendered depth differences; both are label-construction choices using the simulator's ground truth, not instances where the paper predicts its own inputs. No load-bearing self-citation chain appears: references to Diffusion Policy, 3D Diffusion Policy, and similar works are external context, and the reported competition result is an external benchmark outcome. The one notable issue is an evaluation-protocol concern, not circularity: Section 3.2 states that the competition's 6 task categories were divided into 8 tasks ("we further divided the place_object_scale category into three separate tasks"), and the paper reports only a single aggregate success rate without per-task breakdowns or episode counts. This makes the headline number harder to verify against the official protocol, but it does not make the derivation circular. Accordingly, no circular step is identified, and the circularity score is 0.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The central claim rests on simulator-generated labels, keypose tagging from the RoboTwin expert, correct language-based task routing, and the Markov assumption of single-frame observations; none of these are validated by ablations in the report. There are eight hand-chosen hyperparameters or design choices listed above, and no new physical entities are introduced.

free parameters (8)
  • Prediction horizon H = 8
    The diffusion U-Net predicts H=8 future action anchors; chosen by hand with no ablation on H (Section 3.3, 'We use H=8 in AnchorDP3').
  • FPS point count N = 4096
    Input point cloud is downsampled to N=4096 points to fit a 24GB GPU; no sensitivity analysis is reported (Section 3.2, Down sampling).
  • DAgger collection probability = 10%
    There is a 10% probability of collecting wrong-pose observations during the render phase; no ablation of this rate is given (Section 3.4, Data aggregation).
  • Random frames rendered between keyposes = 2
    Only two randomly chosen frames per interval between keypoints are rendered; no ablation shows that 2 is optimal (Section 3.4, Sparse trajectory dataset construction).
  • Task feature embedding and MLP hidden dimensions = 192 / 128
    The per-task encoders use a 128-dim hidden MLP and a 192-dim output embedding; chosen by design with no search reported (Section 3.2, Task-conditioned observation encoders).
  • Depth-difference mask threshold delta = 1e-5
    Pixels with depth difference below 1e-5 are labeled as critical-object locations; the threshold is a hand-set tolerance (Section 3.2, Simulator-supervised semantic segmentation).
  • BERT instruction training set size = 10,000
    The language classifier is fine-tuned on 10,000 paraphrased instructions; no classifier accuracy or saturation analysis is reported (Section 3.2, Language instruction following).
  • Task taxonomy split = 8 tasks (from 6 official categories)
    The authors split place_object_scale into three tasks because of shape variation; this changes the evaluation space and the meaning of the aggregate success rate (Section 3.2).
assumptions (6)
  • domain assumption Simulator rendering and ground-truth object lists yield correct semantic masks.
    Section 3.2 obtains masks by rendering the scene with task-critical objects hidden; if lighting, transparency, or depth noise break the depth-difference assumption, the segmentation labels are wrong and the affordance prior degrades.
  • domain assumption RoboTwin's expert policy atomic operations define sufficient keypose anchors.
    Section 3.3 tags the terminal action of each atomic module (MoveToPose, SetGripper) as a keypose and assumes these are the decision points needed for successful imitation; no experiment verifies anchor completeness.
  • domain assumption The fine-tuned BERT classifier selects the correct task encoder.
    Section 3.2 routes the point cloud through the encoder chosen by arg max over BERT logits; a wrong task choice would feed mismatched features into the shared action expert. No classifier accuracy is reported.
  • domain assumption A single current observation is a sufficient Markov state for the policy.
    Section 3.3 states the action expert uses only one frame of observation with no history; if velocities or temporal context matter, the closed-loop performance would degrade. No ablation with observation history is given.
  • standard math Diffusion denoising, FiLM conditioning, and PointNet++ behavior follow prior work.
    The method inherits the generative model from references [5,6,25,26] and conditioning from [27] without re-derivation; these are standard, published tools.
  • domain assumption The 8-task aggregate mean is representative of overall performance.
    The paper reports one average success rate over a self-modified task split with no per-task or variance statistics; the representativeness of the aggregate is assumed, not shown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AnchorDP3: 3D Affordance Guided Sparse Diffusion Policy for Robotic Manipulation." pith.science (2026). https://pith.science/paper/7KMITY63

@misc{pith2026250619269,
  author       = {Pith},
  title        = {Pith review of: AnchorDP3: 3D Affordance Guided Sparse Diffusion Policy for Robotic Manipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7KMITY63}},
  note         = {Machine review of arXiv:2506.19269}
}
read the original abstract

We present AnchorDP3, a diffusion policy framework for dual-arm robotic manipulation that achieves state-of-the-art performance in highly randomized environments. AnchorDP3 integrates three key innovations: (1) Simulator-Supervised Semantic Segmentation, using rendered ground truth to explicitly segment task-critical objects within the point cloud, which provides strong affordance priors; (2) Task-Conditioned Feature Encoders, lightweight modules processing augmented point clouds per task, enabling efficient multi-task learning through a shared diffusion-based action expert; (3) Affordance-Anchored Keypose Diffusion with Full State Supervision, replacing dense trajectory prediction with sparse, geometrically meaningful action anchors, i.e., keyposes such as pre-grasp pose, grasp pose directly anchored to affordances, drastically simplifying the prediction space; the action expert is forced to predict both robot joint angles and end-effector poses simultaneously, which exploits geometric consistency to accelerate convergence and boost accuracy. Trained on large-scale, procedurally generated simulation data, AnchorDP3 achieves a 98.7% average success rate in the RoboTwin benchmark across diverse tasks under extreme randomization of objects, clutter, table height, lighting, and backgrounds. This framework, when integrated with the RoboTwin real-to-sim pipeline, has the potential to enable fully autonomous generation of deployable visuomotor policies from only scene and instruction, totally eliminating human demonstrations from learning manipulation skills.

Figures

Figures reproduced from arXiv: 2506.19269 by the authors.

Figure 1
Figure 1. Examples of 3D simulation tasks generated by the RoboTwin system, including diverse object configurations and ma￾nipulation goals across multiple environments. To address these challenges, we pro￾pose AnchorDP3, a diffusion-policy￾based framework for dual-arm robotic manipulation under high scene vari￾ability. It comprises three key compo￾nents, each aligned with specific bot￾tlenecks identified in the RoboTwin Chal… view at source ↗
Figure 2
Figure 2. The overall workflow of AnchorDP3. We use both the joint space vector (qpos) and end effector pose to represent the robot state and action. Note that the rotation is not represented through Euler Angles or quaternions. Instead, we use the first two columns of the rotation matrix R3×3 to construct a continuous representation of rotation, which is beneficial for training neural networks [24] [PITH_FULL_IMAGE:figures/… view at source ↗
Figure 3
Figure 3. Simulator-Supervised Semantic Segmentation. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: RoboTwin dataset construction based on action anchors. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Sign in 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. AffordTrajDP: Dynamic Affordance-Guided Visuomotor Policy Learning for Robotic Manipulation

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A robot policy that propagates a retrieved contact point through predicted object poses, producing a time-varying affordance trajectory, improves manipulation success over static affordance and pose-only baselines.

Reference graph

Works this paper leans on

26 extracted references · 12 canonical work pages · cited by 1 Pith paper

  1. [1]

    Robotwin 2.0: A scalable data generator and benchmark with strong domain randomization for robust bimanual robotic manipulation.arXiv preprint arXiv:2506.18088, 2025

    Tianxing Chen, Zanxin Chen, Baijun Chen, Zijian Cai, Yibin Liu, Qiwei Liang, Zixuan Li, Xianliang Lin, Yiheng Ge, Zhenyu Gu, Weiliang Deng, Yubin Guo, Tian Nian, Xuanbing Xie, Qiangyu Chen, Kailun Su, Tianling Xu, Guodong Liu, Mengkang Hu, Huan ang Gao, Kaixuan Wang, Zhixuan Liang, Yusen Qin, Xiaokang Yang, Ping Luo, and Yao Mu. Robotwin 2.0: A scalable d...

    arXiv 2025

  2. [2]

    Mimicgen: A data generation system for scalable robot learning using human demonstrations.arXiv preprint arXiv:2310.17596, 2023

    Ajay Mandlekar, Soroush Nasiriany, Bowen Wen, Iretiayo Akinola, Yashraj Narang, Linxi Fan, Yuke Zhu, and Dieter Fox. Mimicgen: A data generation system for scalable robot learning using human demonstrations.arXiv preprint arXiv:2310.17596, 2023

    arXiv

  3. [3]

    Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  4. [4]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  5. [5]

    Diffusion policy: Visuomotor policy learning via action diffusion.The International Journal of Robotics Research, page 02783649241273668, 2023

    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.The International Journal of Robotics Research, page 02783649241273668, 2023

    2023

  6. [7]

    Diffusion policy policy optimization.arXiv preprint arXiv:2409.00588, 2024

    Allen Z Ren, Justin Lidard, Lars L Ankile, Anthony Simeonov, Pulkit Agrawal, Anirudha Majumdar, Benjamin Burchfiel, Hongkai Dai, and Max Simchowitz. Diffusion policy policy optimization.arXiv preprint arXiv:2409.00588, 2024

    arXiv 2024

  7. [8]

    Equivariant diffusion policy.arXiv preprint arXiv:2407.01812, 2024

    Dian Wang, Stephen Hart, David Surovik, Tarik Kelestemur, Haojie Huang, Haibo Zhao, Mark Yeat- man, Jiuguang Wang, Robin Walters, and Robert Platt. Equivariant diffusion policy.arXiv preprint arXiv:2407.01812, 2024

    arXiv 2024

  8. [9]

    The theory of affordances:(1979)

    James J Gibson. The theory of affordances:(1979). InThe people, place, and space reader, pages 56–60. Routledge, 2014

    work page 1979

Show all 26 references
  1. [10]

    Affordancenet: An end-to-end deep learning approach for object affordance detection

    Thanh-Toan Do, Anh Nguyen, and Ian Reid. Affordancenet: An end-to-end deep learning approach for object affordance detection. In2018 IEEE international conference on robotics and automation (ICRA), pages 5882–5889. IEEE, 2018

    2018

  2. [11]

    Phrase-based affordance detection via cyclic bilateral interaction.IEEE Transactions on Artificial Intelligence, 4(5):1186–1198, 2022

    Liangsheng Lu, Wei Zhai, Hongchen Luo, Yu Kang, and Yang Cao. Phrase-based affordance detection via cyclic bilateral interaction.IEEE Transactions on Artificial Intelligence, 4(5):1186–1198, 2022

    2022

  3. [12]

    Affordance grounding from demon- stration video to target image

    Joya Chen, Difei Gao, Kevin Qinghong Lin, and Mike Zheng Shou. Affordance grounding from demon- stration video to target image. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6799–6808, 2023

    2023

  4. [13]

    Locate: Localize and transfer object parts for weakly supervised affordance grounding

    Gen Li, Varun Jampani, Deqing Sun, and Laura Sevilla-Lara. Locate: Localize and transfer object parts for weakly supervised affordance grounding. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10922–10931, 2023

    2023

  5. [14]

    Learning affordance grounding from exocentric images

    Hongchen Luo, Wei Zhai, Jing Zhang, Yang Cao, and Dacheng Tao. Learning affordance grounding from exocentric images. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2252–2261, 2022

    2022

  6. [15]

    Grounded human-object interaction hotspots from video

    Tushar Nagarajan, Christoph Feichtenhofer, and Kristen Grauman. Grounded human-object interaction hotspots from video. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 8688–8697, 2019

    2019

  7. [16]

    3d affordancenet: A benchmark for visual object affordance understanding

    Shengheng Deng, Xun Xu, Chaozheng Wu, Ke Chen, and Kui Jia. 3d affordancenet: A benchmark for visual object affordance understanding. Inproceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1778–1787, 2021

    2021

  8. [17]

    Grounding 3d object affordance from 2d interactions in images

    Yuhang Yang, Wei Zhai, Hongchen Luo, Yang Cao, Jiebo Luo, and Zheng-Jun Zha. Grounding 3d object affordance from 2d interactions in images. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 10905–10915, 2023. 9

    2023

  9. [18]

    Open- vocabulary affordance detection in 3d point clouds

    Toan Nguyen, Minh Nhat Vu, An Vuong, Dzung Nguyen, Thieu V o, Ngan Le, and Anh Nguyen. Open- vocabulary affordance detection in 3d point clouds. In2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 5692–5698. IEEE, 2023

    2023

  10. [19]

    Rt-1: Robotics transformer for real-world control at scale.arXiv preprint arXiv:2212.06817, 2022

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen 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

  11. [20]

    Gr-2: A generative video-language-action model with web-scale knowledge for robot manipulation.arXiv preprint arXiv:2410.06158, 2024

    Chi-Lam Cheang, Guangzeng Chen, Ya Jing, Tao Kong, Hang Li, Yifeng Li, Yuxiao Liu, Hongtao Wu, Jiafeng Xu, Yichu Yang, et al. Gr-2: A generative video-language-action model with web-scale knowledge for robot manipulation.arXiv preprint arXiv:2410.06158, 2024

    arXiv

  12. [21]

    π0: A vision-language-action flow model for general robot control, 2024.URL https://arxiv

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

    2024 arXiv

  13. [22]

    Towards generalist robot policies: What matters in building vision-language- action models.arXiv preprint arXiv:2412.14058, 2024

    Xinghang Li, Peiyan Li, Minghuan Liu, Dong Wang, Jirong Liu, Bingyi Kang, Xiao Ma, Tao Kong, Hanbo Zhang, and Huaping Liu. Towards generalist robot policies: What matters in building vision-language- action models.arXiv preprint arXiv:2412.14058, 2024

    2024 arXiv

  14. [23]

    Fast: Efficient action tokenization for vision-language-action models.arXiv preprint arXiv:2501.09747, 2025

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

    2025 arXiv

  15. [24]

    On the continuity of rotation represen- tations in neural networks

    Yi Zhou, Connelly Barnes, Jingwan Lu, Jimei Yang, and Hao Li. On the continuity of rotation represen- tations in neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5745–5753, 2019

    2019

  16. [25]

    Diffusion policy: Visuomotor policy learning via action diffusion.The International Journal of Robotics Research, 2023

    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.The International Journal of Robotics Research, 2023

    2023

  17. [26]

    3d diffusion policy: Generalizable visuomotor policy learning via simple 3d representations.arXiv preprint arXiv:2403.03954, 2024

    Yanjie Ze, Gu Zhang, Kangning Zhang, Chenyuan Hu, Muhan Wang, and Huazhe Xu. 3d diffusion policy: Generalizable visuomotor policy learning via simple 3d representations.arXiv preprint arXiv:2403.03954, 2024

    2024 arXiv

  18. [27]

    Visual reasoning with a general conditioning layer, courville

    E Perez, F Strub, H De Vries, and V Dumoulin. Visual reasoning with a general conditioning layer, courville. InIn Proceedings of the AAAI Conference on Artificial Intelligence, 2017. 10

    2017

Pith tools

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