Pith. sign in

REVIEW 3 major objections 6 minor 48 references

A policy trained only on static object demonstrations can manipulate moving objects by predicting the future pose and preserving the hand-object relative pose, according to SIDO.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-07-31 23:31 UTC pith:GIAX4F4D

load-bearing objection SIDO's core augmentation is new and mostly works, but the paper asserts its mechanism instead of proving it and the 20-rollout tables need error bars before the claim is fully sold. the 3 major comments →

arxiv 2607.27890 v1 pith:GIAX4F4D submitted 2026-07-30 cs.RO

Static In, Dynamic Out: Counterfactual Action Augmentation for Moving Object Manipulation

Woo Chul Shin , Zhenyang Chen , Alfred Cueva , Nadun Ranawaka Arachchige , Yingyan Celine Lin , Benjamin Joffe , Shreyas Kousik , Danfei Xu This is my paper
classification cs.RO
keywords imitation learningvisuomotor policycounterfactual data augmentationmoving object manipulationhand-object relative poseobject pose predictiondiffusion policybehavior cloning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Static In, Dynamic Out (SIDO) claims that a robot policy trained exclusively on demonstrations of stationary objects can manipulate objects that move in ways never seen during training. The key move is to split the problem into predicting where the object will be one action-chunk ahead and learning a goal-conditioned reach that preserves the demonstrated hand-object relative pose. SIDO augments each static demonstration by virtually displacing the object and morphing the action chunk so that the relative pose at the replanning point and chunk terminus is unchanged. At deployment, any object pose predictor supplies the future position, and the policy reaches for it. Experiments across three simulated tasks, five motion patterns, and two real-world platforms report improved moving-object success while keeping static-object performance intact.

Core claim

The paper's central claim is that the temporal gap between static demonstrations and dynamic deployment can be closed by data augmentation, not by collecting dynamic demonstrations. The hand-object relative pose r_t(k) = p_eef_{t+k} - P_{t+k} is the invariant that carries the argument. For each static transition, SIDO samples a displacement δ, writes the counterfactual future object position into the dedicated pose channel, and morphs the demonstrated action chunk so that r_t(Ta) and r_t(Tp) match the expert's relative pose. The resulting goal-conditioned policy, paired with any object pose predictor, tracks a moving object at rollout. In simulation and on two real-world platforms, SIDO impr

What carries the argument

The central object is the hand-object relative pose r_t(k) = p_eef_{t+k} - P_{t+k}, defined for offsets k up to the prediction horizon Tp. The invariant constraint preserves r_t(Ta) and r_t(Tp), where Ta is the replanning point and Tp the chunk terminus, when the object is virtually displaced by δ. The counterfactual augmentation edits only the dedicated object-pose observation channel, leaving images unchanged, and morphs the action chunk via a linear ramp (H-SIDO) or an MPPI refinement under a learned dynamics model (DynaSIDO). At deployment, a plug-and-play object pose predictor fills that same channel with the forecast future position.

Load-bearing premise

The policy must use the dedicated object-pose channel to decide where to reach, not the pixels; otherwise the training-time mismatch between unchanged images and displaced pose signals diverges from deployment, where images and pose channel both point to the future.

What would settle it

Feed a SIDO-trained policy a static image pair but a pose channel displaced to a fake future position; if the gripper follows the fake position, the pose channel is what matters. Then feed the true moving images but hold the pose channel at the current pose; if the gripper lags or misses, the pixel path is not driving the reach. Either outcome would settle the load-bearing assumption.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Manipulation of moving objects no longer requires expensive, unsafe teleoperation on moving targets; large static demonstration datasets become reusable.
  • A single trained policy can be deployed across different motion regimes by swapping the object pose predictor, with no retraining.
  • The invariant constraint at replanning and chunk-terminus horizons prevents the compounding out-of-distribution drift that otherwise breaks behavior cloning on moving objects.
  • Improving the pose predictor directly improves task success, as forecast error and success rate correlate across the tested predictors.
  • The method is policy-agnostic, holding for both diffusion and equivariant diffusion policies, suggesting broad applicability.

Where Pith is reading between the lines

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

  • If the pose-channel reading holds, the same counterfactual trick could be applied to other sparse, dedicated observation channels (e.g., tactile or proprioceptive features) without retraining the full policy.
  • The approximate Galilean equivariance the paper observes hints that training over a range of displacement magnitudes may yield policies that generalize to unseen velocities, not just unseen directions — a testable extension beyond the reported speeds.
  • The collision-ignoring limitation suggests a natural next step: replace the linear or MPPI morph with a collision-aware planner so the displaced reach can navigate clutter, extending SIDO beyond open tabletop scenes.
  • A causal ablation that zeroes or corrupts the pose channel while leaving images intact would directly test the assumption that the policy reads the pose channel; the paper's t-SNE analysis is correlational, not causal.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes SIDO (Static In, Dynamic Out), a counterfactual action augmentation that enables a visuomotor policy trained only on static-object demonstrations to manipulate objects that move at test time. The method takes each static demonstration transition, displaces the object in the dedicated pose channel by a vector δ, and morphs the action chunk so that the hand-object relative pose at the replanning horizon Ta and the prediction horizon Tp is preserved relative to the displaced object. Two variants are presented: H-SIDO, a heuristic linear waypoint ramp, and DynaSIDO, which refines the chunk with MPPI under a learned dynamics model. At deployment, an object pose predictor writes the predicted future object position into the pose channel. Evaluation covers three simulated MimicGen tasks under five motion patterns, two real-world platforms (Gantry, Peachtree), two policy classes (Diffusion Policy and Equivariant Diffusion Policy), and three object pose predictors. The paper claims that SIDO improves moving-object success over baselines while preserving static performance, transfers across policy classes and predictors, and works by preserving the hand-object invariant.

Significance. If the claims hold, this is a practically useful result: it separates moving-object manipulation from expensive moving-object teleoperation and lets the policy be trained once and paired at deployment with any object pose forecaster. The paper has several genuine strengths: the method is clearly specified; the comparison against a policy trained on moving-object demonstrations (Appendix B) is a valuable and often-missing baseline; the evaluation includes two real-world platforms and a speed sweep (Appendix A); and the predictor-swap experiment (Table 4) directly tests the claimed plug-and-play property. The central limitations are evidential: the causal mechanism is asserted rather than verified, the augmentation scale is calibrated to the test speed, and the headline success differences are small relative to the 20-rollout sample size with no confidence intervals or multi-seed results. These issues do not invalidate the idea, but they currently leave the strength of the empirical claims unclear.

major comments (3)
  1. [§4.2, Fig. 5] The sentence 'By construction the policy reads the object’s position from the dedicated observation channel, not from the pixels' is an empirical claim about the learned policy’s information flow, not a property of the architecture or training objective. The counterfactual observations keep images unchanged and edit only the pose channel; this pressures the policy to use the pose channel but does not guarantee it. At deployment, the images show the object at its current position while the pose channel holds the predicted future position. If the policy relies on pixel-level object localization, the training distribution (static image + displaced pose) diverges from deployment (moving image + future pose), and the reported gains cannot be attributed to SIDO’s mechanism. The t-SNE and tracking-slope evidence is consistent with pose-channel use but does not establish causality. Because the v
  2. [Appendix F, Appendix I] The finite-difference predictor is described as stepping 'along the direction of those median differences with magnitude v Tp' (Appendix F), and Appendix I sets v=0.001, Tp=16, giving ||δ||=0.016 simulation units, which matches the 2 cm/s test speed over 0.8 s (1.6 cm). Thus the predictor’s forecast magnitude and the augmentation scale are both calibrated to the same test speed. The main experiments therefore demonstrate generalization over motion direction, not over speed or displacement magnitude. The abstract’s claim of adapting to 'unseen object motion' overstates this. Appendix C confirms that handling varying velocity requires training over a range of displacement magnitudes and using a learned predictor (FERNN), so this is a real limitation. The paper should either evaluate at speeds outside the augmentation range or explicitly qualify the claim as direction generalization at a kn
  3. [Tables 1, 3, 4] Every success cell in the evaluation is based on 20 rollouts, with no error bars, confidence intervals, or multi-seed variance. Several headline differences are within binomial sampling noise: Peachtree dynamic is 0.60 for DP+SIDO versus 0.45 for DP (cur. pose)—12 versus 9 successes out of 20; Gantry random dynamic is 0.25 versus 0.20; and in Table 4 the FERNN predictor’s 0.40 versus 0.25 for the MLP is 8 versus 5 successes. Additionally, Table 1 shows EquiDP+SIDO is worse than EquiDP (cur. pose) on Mug x-axis and random (0.10 vs 0.35 and 0.00 vs 0.15), which the text attributes to a 'gap between the commanded and executed trajectories' without supporting evidence. These issues directly weaken H1 and H3. Reporting binomial confidence intervals, running multiple seeds, and providing per-condition significance tests would substantially strengthen the empirical claims.
minor comments (6)
  1. [Appendix I] The conversion from simulation units to physical centimeters is not stated. The reader cannot verify that ||δ||=0.016 simulation units corresponds to 1.6 cm at 2 cm/s; please add the conversion or report δ in physical units.
  2. [Appendix A/B] There are duplicate table numbers: Appendix A contains a Table 6 (speed sweep) and Appendix B also has a Table 6 (collection cost) and Table 7 (results). Renumber the appendices' tables sequentially.
  3. [§5, H2] The tracking slope β_x is reported as a point estimate from a least-squares fit, but no standard error or R² is given. Adding these would make the comparison between SIDO (0.42) and the baselines more interpretable.
  4. [§4.2] The phrase 'By construction the policy reads...' should be reworded, since it is a design assumption rather than an architectural guarantee. This also aligns with major comment 1.
  5. [§5] The statement that DP+AC 'falls to 0.00, below even plain DP' is interesting but the mechanism ('the controller smooths it away') is not demonstrated. A plot of the executed versus commanded trajectory for DP+AC would help.
  6. [Appendix C] The closing remark about an 'approximate Galilean equivariance effect' is vague and not formalized. Either define the symmetry precisely or remove the remark.

Circularity Check

1 steps flagged

Moving-object 'prediction' is partly matched to the test condition: the augmentation displacement δ equals the object's per-chunk motion at the 2 cm/s test speed.

specific steps
  1. fitted input called prediction [§4.2 Counterfactual Augmentation; Appendix I Implementation Details; §5 Experiments (simulation speed)]
    "we draw a uniform direction θ∼ U[0,2π) and set δ=v T p [cosθ,sinθ,0] ⊤, where v is a free parameter of the augmentation controlling the displacement magnitude ... v= 0.001simulation units/step,∥δ∥= 0.016simulation units ... the object moving at2cm/s at test time ... δcorresponding to the 2cm/s deployment speed"

    The counterfactual displacement magnitude is set so that δ equals the object's actual displacement over the prediction horizon at the single test speed used in the main experiments (2 cm/s × 0.8 s = 1.6 cm = 0.016 sim units). Therefore the training-time future pose that SIDO conditions on is the same future pose the policy is asked to reach during deployment. The reported moving-object success is not a prediction from static demonstrations alone; it is partly attained because the test-speed value is encoded into the free parameter v. Random uniform directions and the separate predictor provide independent content, but the main empirical claim is matched to the tested speed by construction.

full rationale

The paper's central factorization—future-pose prediction plus goal-conditioned reaching—is not circular per se: Eq. (2) preserves the hand-object relative pose at Ta and Tp, and the policy is trained on augmented static demonstrations and evaluated with a swapped predictor. The main circularity concern is the displacement scale v: §4.2 defines δ = v Tp in a uniform direction, and Appendix I sets v = 0.001, ∥δ∥ = 0.016, while §5 tests objects moving at 2 cm/s. Since Tp = 16 steps at 20 Hz is 0.8 s, δ is exactly the object's per-chunk motion at the tested speed. Thus the counterfactual future position is the deployment future position, so part of the moving-object success is a matched-condition result rather than an out-of-distribution prediction. The paper's assertion in §4.2 that 'the policy reads the object's position from the dedicated observation channel, not from the pixels' is load-bearing but is supported only by t-SNE and tracking-slope evidence, not a causal ablation; that is an unsupported empirical claim, not a circularity. Self-citations [8,15,16,17] appear only in related work and are not load-bearing. Overall score 4: one matched-parameter input partially compromises the 'unseen motion' claim, but random directions, separate predictors, and speed sweeps give the method independent content.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claim rests on an explicit geometric prior (hand-object invariant), a displacement scale matched to deployment speed, and an assumption about which observation channel the policy uses. No new physical entities are introduced.

free parameters (5)
  • Displacement scale v = 0.001 simulation units/step (∥δ∥=0.016 over Tp=16)
    Controls the counterfactual displacement magnitude; set so δ matches the 2 cm/s test object motion over the prediction horizon (Appendix I, §4.2).
  • Augmentation ratio α = 0.2 static / 0.8 counterfactual
    Probability of returning the original sample; a hyperparameter preserving static performance (Appendix I).
  • Prediction horizon Tp / action horizon Ta = Tp=16, Ta=8
    Chunking hyperparameters; Tp is also the look-ahead of the pose predictor (Appendix I).
  • MPPI search settings = 128 samples, 10 iterations, β=0.5
    Optimizer hyperparameters for DynaSIDO refinement (Appendix I).
  • Object pose predictor training data = 5 minutes of moving-object trajectories
    Learned predictors (MLP, FERNN) are trained on deployment-like trajectories, a data input not required for the policy itself (Appendix F).
axioms (5)
  • domain assumption The policy reads the object's position from the dedicated pose channel, not from the pixels (§4.2: 'By construction the policy reads the object’s position from the dedicated observation channel, not from the pixels').
    Counterfactual observations leave images unchanged and only edit the pose channel; if the policy uses pixels for object localization, the augmented training distribution does not match deployment.
  • ad hoc to paper Preserving the hand-object relative pose at Ta and Tp is sufficient to generate successful counterfactual actions.
    The invariant constraint is the core design choice; no proof that two-point matching is sufficient for arbitrary objects and motions.
  • domain assumption Object motion is planar (δ_z=0) for the counterfactual displacement.
    All test motion patterns are in the XY plane (§4.2, Appendix E); out-of-plane motion is not handled.
  • domain assumption Demonstrations are static during the pre-grasp window (P_t = P_static for t ≤ T_g).
    This is the premise of the problem statement (§3), true for collected data but restricts the method's input.
  • domain assumption The learned dynamics model f_ψ captures the closed-loop controller's behavior.
    DynaSIDO uses f_ψ rolled out for MPPI refinement (§4.2); if f_ψ is inaccurate, the refined chunks may not track as intended.

pith-pipeline@v1.3.0-daily-deepseek · 160 in / 10688 out tokens · 147794 ms · 2026-07-31T23:31:12.992661+00:00 · methodology

0 comments
read the original abstract

Visuomotor policies have advanced on manipulation tasks where the target object stays static during execution, but real deployments break this assumption: parts drift on conveyors and fruits sway in the wind. We introduce Static In, Dynamic Out (SIDO), a counterfactual action augmentation that enables a policy trained only on static object demonstrations to adapt to unseen object motion at test time. Our key idea is to factorize moving object manipulation into two sub-problems: predicting where the object will be, and reaching that predicted pose. SIDO displaces the object to a counterfactual future position and morphs the demonstrated action chunk to preserve the hand-object relative pose, yielding a goal-conditioned policy. At deployment an object pose predictor supplies the future position. Across three simulated tasks (Mug, Square, Stack) under five object motion patterns and two real-world tasks (Gantry, Peachtree), SIDO improves moving object success over the baselines while preserving static object performance. Project website: https://sido-staticindynamicout.github.io/.

Figures

Figures reproduced from arXiv: 2607.27890 by Alfred Cueva, Benjamin Joffe, Danfei Xu, Nadun Ranawaka Arachchige, Shreyas Kousik, Woo Chul Shin, Yingyan Celine Lin, Zhenyang Chen.

Figure 1
Figure 1. Figure 1: Static In, Dynamic Out (SIDO) enables a policy trained only on static object demon￾strations to manipulate moving objects. At training time, SIDO creates counterfactual data pairs by virtually displacing the object and morphing the expert action to preserve the hand-object relative pose, and trains a goal-conditioned policy on them. At deployment, an object pose predictor supplies the object’s future pose,… view at source ↗
Figure 2
Figure 2. Figure 2: SIDO overview. Goal-conditioned policy: from each static demonstration, SIDO (1) fits a robot dynamics model fψ on expert data, (2) virtually displaces the object, and (3) morphs the expert chunk into a counterfactual chunk a˜t:t+Tp−1 that preserves the hand-object relative pose at Ta and Tp. H-SIDO uses a heuristic ramp; DynaSIDO refines it under fψ. Object pose predictor: SIDO pairs plug-and-play with an… view at source ↗
Figure 3
Figure 3. Figure 3: Object motion patterns across simulation and real-world tasks. Left: simulation tasks under five object motion patterns (static, x-axis, y-axis, circle, random). Middle: real-world Gantry under four object motion patterns (static, x-axis, y-axis, random). Right: real-world Peachtree under two object motion patterns (static, dynamic). waypoint directly, a˜t+k = at+k + ρk δ, ρk = min(k, Ta − 1) Ta − 1 , k ∈ … view at source ↗
Figure 4
Figure 4. Figure 4: Gripper tracks the object under SIDO, baselines lag behind. World-frame x-z scatter of commanded EEF at action terminus Ta (circles) and object position (squares) on Gantry x-axis. SIDO’s commands sweep with the object, preserving the hand-object relative pose, baselines do not. Right: tracking slope βx of commanded EEF on observed object x, 20 rollouts. H3: SIDO transfers across policy classes and predict… view at source ↗
Figure 5
Figure 5. Figure 5: SIDO’s wrist view stays in-distribution with the static training data. t-SNE of pre￾grasp wrist frames in each method’s fine-tuned DINOv3 space, 20 rollouts per method on Gantry x-axis. Bottom row: wrist view at the grasp-attempt moment. SIDO’s frame matches the dataset reference, while baselines drift away and miss or off-center the object. SIDO can also pair with a plug-and-play predictor without retrain… view at source ↗
Figure 6
Figure 6. Figure 6: H-SIDO vs. DynaSIDO success rate (↑) at higher object speeds on Stack x-axis, 20 rollouts per cell. The two match at and below the 2 cm/s deployment speed, and DynaSIDO pulls ahead as the object moves faster. B SIDO vs Training on Moving Object Demonstrations This section compares SIDO against the coupled alternative, a single policy trained directly on moving object demonstrations. Setup. The moving objec… view at source ↗
Figure 7
Figure 7. Figure 7: Static vs moving object demonstration coverage. Moving object reaches only 80.02% of the workspace extent the static 8 × 8 grid covers. The lower static performance follows from the anticipatory offset present in moving object chunks. The policy learns to reach ahead of the object’s current position by roughly its displacement over the chunk horizon. On a static object, this offset has no motion to track, … view at source ↗
Figure 8
Figure 8. Figure 8: SIDO real-world evaluation system. Both platforms use a Franka arm with the shared perception and gripper stack shown in the center: a ZED 2 for the agent view, a ZED Mini for the wrist view, and a Robotiq 2F-85 parallel-jaw gripper. Gantry (left): a low-cost LCLC gantry translates a capsule, which the arm picks and places into a box. Peachtree (right): an AgileX Tracer mobile base sways an artificial peac… view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of object pose predictor forecasts on Gantry random. Each panel shows pre-grasp snapshots in the world frame: the gray dashed line traces the object’s recent pose history, the light green marks the current object pose and the dark green the ground-truth future pose one horizon Tp ahead. The three predictors forecast that future pose from the history, finite-difference (red), MLP (blue), and F… view at source ↗

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

48 extracted references · 32 linked inside Pith

  1. [1]

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song. Diffusion policy: Visuomotor policy learning via action diffusion, 2024. URLhttps://arxiv.org/ abs/2303.04137

    Pith/arXiv arXiv 2024

  2. [2]

    T. Z. Zhao, V . Kumar, S. Levine, and C. Finn. Learning fine-grained bimanual manipulation with low-cost hardware, 2023. URLhttps://arxiv.org/abs/2304.13705

    Pith/arXiv arXiv 2023

  3. [3]

    Reuss, M

    M. Reuss, M. Li, X. Jia, and R. Lioutikov. Goal-Conditioned Imitation Learning using Score- based Diffusion Policies. InProceedings of Robotics: Science and Systems, Daegu, Republic of Korea, 7 2023. doi:10.15607/RSS.2023.XIX.028

    doi:10.15607/rss.2023.xix.028 2023

  4. [4]

    Collaboration, A

    E. Collaboration, A. O’Neill, A. Rehman, A. Gupta, A. Maddukuri, A. Gupta, A. Padalkar, A. Lee, A. Pooley, A. Gupta, A. Mandlekar, A. Jain, A. Tung, A. Bewley, A. Herzog, A. Ir- pan, A. Khazatsky, A. Rai, A. Gupta, A. Wang, A. Kolobov, A. Singh, A. Garg, A. Kembhavi, A. Xie, A. Brohan, A. Raffin, A. Sharma, A. Yavary, A. Jain, A. Balakrishna, A. Wahid, B....

    Pith/arXiv arXiv 2025

  5. [5]

    Khazatsky, K

    A. Khazatsky, K. Pertsch, S. Nair, A. Balakrishna, S. Dasari, S. Karamcheti, S. Nasiriany, M. K. Srirama, L. Y . Chen, K. Ellis, P. D. Fagan, J. Hejna, M. Itkina, M. Lepert, Y . J. Ma, P. T. Miller, J. Wu, S. Belkhale, S. Dass, H. Ha, A. Jain, A. Lee, Y . Lee, M. Memmel, S. Park, I. Radosavovic, K. Wang, A. Zhan, K. Black, C. Chi, K. B. Hatch, S. Lin, J. ...

    Pith/arXiv arXiv 2025

  6. [6]

    H. Xie, B. Wen, J. Zheng, Z. Chen, F. Hong, H. Diao, and Z. Liu. Dynamicvla: A vision- language-action model for dynamic object manipulation, 2026. URLhttps://arxiv.org/ abs/2601.22153

    arXiv 2026

  7. [7]

    Black, M

    K. Black, M. Y . Galliker, and S. Levine. Real-time execution of action chunking flow policies,

  8. [8]

    N. R. Arachchige, Z. Chen, W. Jung, W. C. Shin, R. Bansal, P. Barroso, Y . H. He, Y . C. Lin, B. Joffe, S. Kousik, and D. Xu. Sail: Faster-than-demonstration execution of imitation learning policies, 2025. URLhttps://arxiv.org/abs/2506.11948

    Pith/arXiv arXiv 2025

  9. [9]

    Zhang, T

    Y . Zhang, T. Liang, Z. Chen, Y . Ze, and H. Xu. Catch it! learning to catch in flight with mobile dexterous hands, 2024. URLhttps://arxiv.org/abs/2409.10319

    Pith/arXiv arXiv 2024

  10. [10]

    H. Xue, J. Ren, W. Chen, G. Zhang, Y . Fang, G. Gu, H. Xu, and C. Lu. Reactive diffu- sion policy: Slow-fast visual-tactile policy learning for contact-rich manipulation, 2025. URL https://arxiv.org/abs/2503.02881

    Pith/arXiv arXiv 2025

  11. [11]

    Ameperosa, J

    E. Ameperosa, J. A. Collins, M. Jain, and A. Garg. Rocoda: Counterfactual data augmentation for data-efficient robot learning from demonstrations, 2025. URLhttps://arxiv.org/abs/ 2411.16959

    Pith/arXiv arXiv 2025

  12. [12]

    Glossop, W

    C. Glossop, W. Chen, A. Bhorkar, D. Shah, and S. Levine. Cast: Counterfactual labels improve instruction following in vision-language-action models, 2025. URLhttps://arxiv.org/ abs/2508.13446

    Pith/arXiv arXiv 2025

  13. [13]

    D. Wang, S. Hart, D. Surovik, T. Kelestemur, H. Huang, H. Zhao, M. Yeatman, J. Wang, R. Walters, and R. Platt. Equivariant diffusion policy, 2024. URLhttps://arxiv.org/ abs/2407.01812

    Pith/arXiv arXiv 2024

  14. [14]

    X. Zhu, F. Wang, R. Walters, and J. Shi. Se(3)-equivariant diffusion policy in spherical fourier space, 2025. URLhttps://arxiv.org/abs/2507.01723

    Pith/arXiv arXiv 2025

  15. [15]

    Saxena, M

    V . Saxena, M. Bronars, N. R. Arachchige, K. Wang, W. C. Shin, S. Nasiriany, A. Mandlekar, and D. Xu. What matters in learning from large-scale datasets for robot manipulation, 2025. URLhttps://arxiv.org/abs/2506.13536

    Pith/arXiv arXiv 2025

  16. [16]

    Y . Liu, W. C. Shin, Y . Han, Z. Chen, H. Ravichandar, and D. Xu. Immimic: Cross-domain imitation from human videos via mapping and interpolation, 2025. URLhttps://arxiv. org/abs/2509.10952

    Pith/arXiv arXiv 2025

  17. [17]

    Y . Liu, S. Cheng, X. Yin, W. C. Shin, A. Cueva, Y . Yang, Z. Chen, C. Zhang, and D. Xu. Egoengine: From egocentric human videos to high-fidelity dexterous robot demonstrations,

  18. [18]

    Y . Han, J. Qiu, L. Bai, Z. Xiao, Z. Zeng, Y . Liu, Z. Yang, S. Jain, W. Ma, J. Fu, Y . Zheng, M. Natarajan, M. Z. Irshad, K. Shaw, M. Gombolay, Z. Kira, and H. Ravichan- dar. Video2sim2real: Full-stack autonomous dexterous skill acquisition from a single human video, 2026. URLhttps://arxiv.org/abs/2606.08828. 11

    Pith/arXiv arXiv 2026

  19. [19]

    S. Ross, G. J. Gordon, and J. A. Bagnell. A reduction of imitation learning and structured prediction to no-regret online learning, 2011. URLhttps://arxiv.org/abs/1011.0686

    Pith/arXiv arXiv 2011

  20. [20]

    Laskey, J

    M. Laskey, J. Lee, R. Fox, A. Dragan, and K. Goldberg. Dart: Noise injection for robust imitation learning, 2017. URLhttps://arxiv.org/abs/1703.09327

    Pith/arXiv arXiv 2017

  21. [21]

    Sun and S

    Z. Sun and S. Song. Latent policy barrier: Learning robust visuomotor policies by staying in-distribution, 2025. URLhttps://arxiv.org/abs/2508.05941

    Pith/arXiv arXiv 2025

  22. [22]

    J. Levy, T. Westenbroek, K. Huang, F. Palafox, P. Yin, S. Omidshafiei, D.-K. Kim, A. Gupta, and D. Fridovich-Keil. Simulation distillation: Pretraining world models in simulation for rapid real-world adaptation, 2026. URLhttps://arxiv.org/abs/2603.15759

    Pith/arXiv arXiv 2026

  23. [23]

    Du and S

    M. Du and S. Song. Dynaguide: Steering diffusion polices with active dynamic guidance,

  24. [24]

    H. Fang, S. Li, S. Wang, X. Xi, D. Liang, and X. Bai. Towards generalizable robotic manipu- lation in dynamic environments, 2026. URLhttps://arxiv.org/abs/2603.15620

    Pith/arXiv arXiv 2026

  25. [25]

    B. Wen, W. Yang, J. Kautz, and S. Birchfield. Foundationpose: Unified 6d pose estimation and tracking of novel objects, 2024. URLhttps://arxiv.org/abs/2312.08344

    Pith/arXiv arXiv 2024

  26. [26]

    URLhttps://arxiv.org/abs/2506.13922

    arXiv

  27. [27]

    R. E. Kalman. A new approach to linear filtering and prediction problems.Transactions of the ASME–Journal of Basic Engineering, 82(Series D):35–45, 1960

    1960

  28. [28]

    Mandlekar, S

    A. Mandlekar, S. Nasiriany, B. Wen, I. Akinola, Y . Narang, L. Fan, Y . Zhu, and D. Fox. Mimicgen: A data generation system for scalable robot learning using human demonstrations,

  29. [29]

    T. A. Keller. Flow equivariant recurrent neural networks, 2025. URLhttps://arxiv.org/ abs/2507.14793

    arXiv 2025

  30. [30]

    Garrett, A

    C. Garrett, A. Mandlekar, B. Wen, and D. Fox. Skillmimicgen: Automated demonstration generation for efficient skill learning and deployment, 2024. URLhttps://arxiv.org/ abs/2410.18907

    Pith/arXiv arXiv 2024

  31. [31]

    L. Ke, Y . Zhang, A. Deshpande, S. Srinivasa, and A. Gupta. Ccil: Continuity-based data augmentation for corrective imitation learning, 2024. URLhttps://arxiv.org/abs/2310. 12972

    2024

  32. [32]

    C. Tie, Y . Chen, R. Wu, B. Dong, Z. Li, C. Gao, and H. Dong. Et-seed: Efficient trajectory- level se(3) equivariant diffusion policy, 2025. URLhttps://arxiv.org/abs/2411.03990

    Pith/arXiv arXiv 2025

  33. [33]

    Z. Xue, S. Deng, Z. Chen, Y . Wang, Z. Yuan, and H. Xu. Demogen: Synthetic demonstration generation for data-efficient visuomotor policy learning, 2025. URLhttps://arxiv.org/ abs/2502.16932

    Pith/arXiv arXiv 2025

  34. [34]

    Williams, A

    G. Williams, A. Aldrich, and E. Theodorou. Model predictive path integral control using co- variance variable importance sampling, 2015. URLhttps://arxiv.org/abs/1509.01149

    Pith/arXiv arXiv 2015

  35. [35]

    Sim ´eoni, H

    O. Sim ´eoni, H. V . V o, M. Seitzer, F. Baldassarre, M. Oquab, C. Jose, V . Khalidov, M. Szafraniec, S. Yi, M. Ramamonjisoa, F. Massa, D. Haziza, L. Wehrstedt, J. Wang, T. Darcet, T. Moutakanni, L. Sentana, C. Roberts, A. Vedaldi, J. Tolan, J. Brandt, C. Cou- prie, J. Mairal, H. J ´egou, P. Labatut, and P. Bojanowski. Dinov3, 2025. URLhttps: //arxiv.org/...

    Pith/arXiv arXiv 2025

  36. [36]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition, 2015. URL https://arxiv.org/abs/1512.03385

    Pith/arXiv arXiv 2015

  37. [37]

    J. Yang, Z. ang Cao, C. Deng, R. Antonova, S. Song, and J. Bohg. Equibot: Sim(3)-equivariant diffusion policy for generalizable and data efficient learning, 2024. URLhttps://arxiv. org/abs/2407.01479

    Pith/arXiv arXiv 2024

  38. [38]

    G. Cesa, L. Lang, and M. Weiler. A program to build E(N)-equivariant steerable CNNs. In International Conference on Learning Representations, 2022. URLhttps://openreview. net/forum?id=WE4qe9xlnQw

    2022

  39. [39]

    Y . Zhu, J. Wong, A. Mandlekar, and R. Mart ´ın-Mart´ın. robosuite: A modular simulation framework and benchmark for robot learning.CoRR, abs/2009.12293, 2020. URLhttps: //arxiv.org/abs/2009.12293

    Pith/arXiv arXiv 2009

  40. [40]

    Mandlekar, D

    A. Mandlekar, D. Xu, J. Wong, S. Nasiriany, C. Wang, R. Kulkarni, L. Fei-Fei, S. Savarese, Y . Zhu, and R. Mart´ın-Mart´ın. What matters in learning from offline human demonstrations for robot manipulation.CoRR, abs/2108.03298, 2021. URLhttps://arxiv.org/abs/2108. 03298

    Pith/arXiv arXiv 2021

  41. [41]

    Weiler and G

    M. Weiler and G. Cesa. General E(2)-Equivariant Steerable CNNs. InConference on Neural Information Processing Systems (NeurIPS), 2019. URLhttps://arxiv.org/abs/1911. 08251

    2019

  42. [42]

    J. Cho. Lowest complexity laser cutter. Hackaday.io, 2020. URLhttps://hackaday.io/ project/169347-lowest-complexity-laser-cutter. Project logs

    2020

  43. [43]

    On the analysis of movement smoothness.Journal of NeuroEngineering and Rehabilitation, 12(1):112, 2015

    Balasubramanian, Sivakumar and Melendez-Calderon, Alejandro and Roby-Brami, Agnes and Burdet, Etienne. On the analysis of movement smoothness.Journal of NeuroEngineering and Rehabilitation, 12(1):112, 2015. doi:10.1186/s12984-015-0090-9. URLhttps://doi.org/ 10.1186/s12984-015-0090-9

    doi:10.1186/s12984-015-0090-9 2015

  44. [44]

    Balasubramanian, A

    S. Balasubramanian, A. Melendez-Calderon, and E. Burdet. A robust and sensitive metric for quantifying movement smoothness.IEEE Transactions on Biomedical Engineering, 59(8): 2126–2136, 2012. doi:10.1109/TBME.2011.2179545. 13 8 Appendix A H-SIDO Approximates DynaSIDO in the Low Speed Regime, DynaSIDO Scales Beyond It We study how the two counterfactual ac...

    arXiv 2012

  45. [45]

    Todorov, T

    E. Todorov, T. Erez, and Y . Tassa. Mujoco: A physics engine for model-based control. In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 5026– 5033, 2012. doi:10.1109/IROS.2012.6386109

    arXiv 2012

  46. [2023]

    URLhttps://arxiv.org/abs/2310.17596

    Pith/arXiv arXiv

  47. [2025]

    URLhttps://arxiv.org/abs/2506.07339

    Pith/arXiv arXiv

  48. [2026]

    URLhttps://arxiv.org/abs/2606.12604

    Pith/arXiv arXiv