Pith. sign in

REVIEW 3 major objections 6 minor 39 references

Robot policies that output continuous B-spline curves finish manipulation tasks much faster than discrete action-chunk policies while keeping strong success rates.

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 · grok-4.5

2026-07-13 01:28 UTC pith:3JHLHPQ3

load-bearing objection Solid CoRL-style engineering: continuous B-spline action heads plus segment alignment actually cut real completion time on Diffusion/ACT pipelines, with the controller-trackability limit already scoped by the authors. the 3 major comments →

arxiv 2607.09648 v1 pith:3JHLHPQ3 submitted 2026-07-10 cs.RO

B-spline Policy: Accelerating Manipulation Policies via B-spline Action Representations

Xiaoshen Han , Haoyu Xiong , Haonan Chen , Chaoqi Liu , Antonio Torralba , Yuke Zhu , Yilun Du This is my paper
classification cs.RO
keywords B-spline policyvisuomotor policyaction representationfast manipulationtemporal rescalingsegment alignmentimitation learningrobot manipulation
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.

Most learned robot policies predict fixed-length lists of discrete future actions. That choice stabilizes long-horizon learning but forces uniform timing and creates jerky joins between successive predictions, so even accurate policies stay slow. This paper argues that if the policy instead predicts the parameters of a continuous B-spline—knots and control points—the identical behavior becomes a smooth curve that can be sampled at high frequency and sped up simply by scaling time. An inference-time alignment step keeps successive curve segments continuous so acceleration does not accumulate boundary jumps. On real cube picking, long-horizon table cleaning, and bimanual cup stacking, and on standard simulation suites, the resulting B-spline Policy cuts completion time substantially while matching or improving success in most moderate-speed settings.

Core claim

Parameterizing visuomotor actions as continuous cubic B-splines, predicted directly by ordinary imitation-learning backbones and executed with temporal rescaling plus segment alignment, substantially reduces task completion time relative to discrete action-chunk baselines while preserving or improving success rates on real and simulated manipulation tasks.

What carries the argument

B-spline Policy (BSP): the policy emits a fixed-size vector of local knots and control points that define a continuous action curve; adaptive fitting turns discrete demonstrations into these targets, and inference-time segment alignment chooses the point on each new segment that best matches the last executed action so consecutive segments join smoothly under acceleration.

Load-bearing premise

The robot’s low-level joint controllers must still track the denser, faster samples of the predicted spline without drifting into states the policy never saw.

What would settle it

On the same real arms and Speed Stacking setup, measure success at 4× temporal scaling with a higher-stiffness controller and perfect segment alignment; if success remains near zero even when tracking error is demonstrably tiny, the continuous-representation claim itself fails rather than the hardware limit.

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

If this is right

  • A trained policy can be sped up at runtime by temporal rescaling without collecting new data or retraining for each target speed.
  • Boundary discontinuities between discrete action chunks cease to be the main obstacle to high-speed visuomotor control.
  • The same continuous head drops into both diffusion and regression (ACT-style) backbones with only an output-space change.
  • Long-horizon tasks gain the largest absolute time savings because smooth high-frequency commands can be sustained across many successive segments.
  • Success holds only while the low-level controller remains able to track the rescaled trajectory.

Where Pith is reading between the lines

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

  • Stiffer, higher-bandwidth joint controllers would likely recover the 4× regime that currently collapses on contact-rich bimanual work.
  • Adaptive knot density already marks which phases need precision; that signal could drive automatic recovery or curriculum decisions.
  • The same knot-and-control-point head could replace discrete action tokens inside vision-language-action models that today emit fixed-length chunks.
  • Because local spline support already encodes multi-step smoothness, policies may need shorter prediction horizons than discrete chunk methods require.

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 introduces B-spline Policy (BSP), an action representation that replaces discrete-time action chunks with continuous cubic B-spline curves parameterized by knots and control points. Demonstrations are converted via adaptive knot insertion (Alg. 1); a standard visuomotor policy (Diffusion or ACT-style regression) then predicts fixed-size local B-spline parameters [U;C]. At inference, segments are temporally rescaled (a_exec(t)=a(nt)), sampled at high frequency, and stitched with an MSE-based segment-alignment step (Eq. 2). Real-world evaluation on Cube Picking, Table Cleaning, and Speed Stacking (ARX5 arms; 20 fixed layouts per method) and simulation on Push-T, RoboMimic, and RoboCasa show reduced average completion times versus discrete chunk baselines and DemoSpeedup, with success rates matched or improved in most 1×–2× settings; aggressive 4× speedup can collapse success when low-level tracking fails.

Significance. Execution speed is a genuine bottleneck for learned visuomotor policies, and the paper cleanly identifies uniform temporal resolution and chunk-boundary discontinuities as algorithmic contributors. Representing actions as B-splines is a natural, compact continuous alternative that integrates with existing Diffusion/ACT pipelines with minimal architectural change, and the inference-time segment alignment mechanism is a practical, falsifiable contribution. The multi-backbone, multi-task real-world results (Table 1), the alignment ablation (Fig. 6), and the Push-T speedup study (Table 2b) provide concrete evidence that continuous spline actions can cut completion time while preserving success inside a controller-feasible regime. Limitations of hardware tracking are stated explicitly (Sec. 6). This is a solid systems/engineering contribution for robot learning venues.

major comments (3)
  1. [Table 1 / Sec. 5.2] Table 1 and Sec. 5.2: Success is reported as raw fractions over 20 fixed layouts with no binomial confidence intervals, standard errors, or hypothesis tests. Several central “preserves success” claims rest on small absolute differences (e.g., Table Cleaning Reg. 4×: 14/20 vs 13/20; Cube Picking many 19–20/20 ties). With n=20 these differences are underpowered; the paper should report CIs (or bootstrap intervals) and, where it claims improvement or parity, a simple test so readers can judge whether the success–time trade-off is statistically supported.
  2. [Sec. 4 / Sec. 5.2 Finding 2 / Sec. 6] Sec. 4 (temporal rescaling a_exec(t)=a(nt)), Sec. 5.2 Finding 2, and Sec. 6: The central speedup claim is conditioned on low-level joint controllers tracking denser, faster spline samples without leaving the policy’s training distribution. The paper correctly shows the failure mode (Speed Stacking Reg.+BSP 4×: 0/20) and attributes it to ARX5 stiffness/accuracy, but does not quantify tracking error, lag, or a per-task maximum reliable speedup. A short characterization (e.g., measured joint tracking RMSE vs. n, or the largest n that keeps success within a stated band of the 1× baseline) would make the feasible regime precise rather than anecdotal.
  3. [Sec. 5.1 / Appendix A.1 DemoSpeedup] Sec. 5.1 baselines and Appendix A.1: Naive n× chunk execution and DemoSpeedup are the right comparisons, but DemoSpeedup’s phase downsampling rates are hand-chosen per task (precision phase fixed at 1; non-precision 4 or 2) after observing that ≥2 on precision “prevents reliable completion.” This makes the DemoSpeedup column a best-effort rather than a fixed protocol. Either freeze the hyperparameters a priori or report a small sensitivity sweep so the comparison cannot be read as tuned against BSP.
minor comments (6)
  1. [Fig. 1 / Fig. 3] Fig. 1 and Fig. 3: The “3.76X faster” callout and the 10 Hz → 100 Hz pipeline are clear, but the figures do not state the corresponding baseline wall-clock time or which task/backbone the 3.76× refers to; add a caption note.
  2. [Eq. (1) / Sec. 3–4] Eq. (1) and Sec. 3: Degree p is fixed to cubic in experiments; state explicitly whether open/clamped knot vectors are used and how endpoint multiplicity is handled when appending repeated final control points for fixed-size targets.
  3. [Appendix A.2 / Alg. 1] Appendix Table 3 (fitting ε ablation) is only on Push-T. A one-sentence note on whether real-world tasks used the same ε=0.002 and whether success was sensitive to it would strengthen reproducibility.
  4. [Sec. 2] Related work (Sec. 2): BEAST and DMPs are cited; a short explicit contrast—BSP as a control-oriented continuous action head + execution pipeline versus tokenizer or motion-primitive encoding—would help readers place the contribution.
  5. [Sec. 3 / Alg. 2] Minor wording: “stablizes” → “stabilizes” (Sec. 3); “segmentment alignment” → “segment alignment” (Alg. 2 caption); arXiv id in the header is 2607.09648 while the abstract footer shows the same—ensure consistency with the public listing.
  6. [Appendix A.1 Controller] Simulation: segment alignment is disabled for delta actions “because it does not have a clear physical interpretation.” A brief alternative (e.g., velocity-consistent stitching) or an explicit statement that sim speedups therefore isolate only the continuous representation would avoid confusion when comparing Table 2b to real-world results.

Circularity Check

0 steps flagged

No significant circularity: empirical action-representation paper whose speedup and success claims are measured on held-out rollouts, not forced by the B-spline fit or self-citation.

full rationale

The paper proposes replacing discrete action chunks with continuous cubic B-spline segments (knots U and control points C) obtained by adaptive FITPACK-style fitting of demonstration trajectories (Alg. 1, tolerance ε). A standard imitation-learning backbone then predicts fixed-size local [U;C] parameters; at inference the curve is temporally rescaled a_exec(t)=a(nt) and optionally segment-aligned by argmin MSE (Eq. 2). These steps are operational definitions of the representation and controller interface, not mathematical derivations that recover a fitted quantity as a “prediction.” Success rates and wall-clock completion times are obtained from independent real-world and simulated rollouts (Tables 1–2) against discrete-chunk baselines; reconstruction error of the spline fit is never used as a performance metric. Related-work citations (BEAST, DMPs, classical spline literature) supply background, not load-bearing uniqueness theorems or self-referential ansätze. Consequently the central empirical claims do not reduce by construction to their inputs, and the circularity score is zero.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The work rests on classical B-spline approximation and standard imitation-learning assumptions, plus several hand-chosen representation and execution hyperparameters (segment size, fitting tolerance, speedup, alignment window). No new physical entities; the invented pieces are methodological (BSP action head and segment alignment). Claims are empirical, so free parameters affect compression and stability more than a closed-form derivation.

free parameters (6)
  • fitting tolerance ε
    Controls adaptive knot insertion and reconstruction error; set to 0.002 for most tasks and 1 for Push-T. Directly affects compression ratio and how much demo detail is retained (Appendix A.2 Table 3).
  • fixed segment size (16 knots / control points)
    Hand-chosen policy output dimension including boundary-support knots; determines local horizon capacity and training target size (Appendix A.1).
  • B-spline degree p
    Fixed to cubic (p=3) for C2 smoothness; not learned from data (Sec. 3).
  • execution speedup factor n (1×/2×/4×)
    Chosen evaluation settings for temporal rescaling a_exec(t)=a(nt); central speed claims depend on these discrete operating points (Sec. 4–5).
  • segment alignment search window λ T_inf
    Hand-set robustness window for matching new segment to last executed action; implemented as u_max(S)/2 in real experiments (Sec. 4 Eq. 2; Appendix A.1).
  • DemoSpeedup phase downsampling rates
    Baseline hyperparameters (precision phase 1; non-precision 4 or 2) chosen by authors after finding ≥2 precision downsampling fails (Appendix A.1).
axioms (5)
  • standard math Cubic B-splines with enough knots can approximate continuous robot trajectories within a chosen error bound and yield continuous velocity/acceleration.
    Invoked via classical spline approximation (De Boor, FITPACK) in Sec. 3 and Algorithm 1.
  • domain assumption Human/teleop demonstration trajectories are a valid training distribution for imitation policies that will be executed faster than demonstrated.
    Standard IL assumption underlying all training; speedup is applied only at inference (Sec. 3–4).
  • domain assumption Geometric shape of a predicted action curve can be preserved under pure temporal rescaling for the tasks studied.
    Stated as a_exec(t)=a(nt) in Sec. 3–4; contact dynamics and controller lag can violate this at high n.
  • ad hoc to paper A fixed-size local knot/control-point vector is a sufficient action representation for standard Diffusion/ACT-style architectures with minimal modification.
    Design choice in Sec. 4 'Fixing the Policy Output Size' enabling plug-and-play training.
  • ad hoc to paper Inference-time MSE alignment of consecutive spline segments sufficiently removes boundary discontinuities for stable high-speed control.
    Core of the proposed execution pipeline (Sec. 4 Eq. 2); ablation in Fig. 6 supports necessity but not universality.
invented entities (2)
  • B-spline Policy (BSP) action head no independent evidence
    purpose: Replace discrete action-chunk outputs with predicted local B-spline parameters for continuous, rescalable control.
    Methodological construct defined in Sec. 3–4; evaluated empirically, not a physical entity. independent_evidence false beyond this paper’s experiments.
  • Inference-time segment alignment mechanism no independent evidence
    purpose: Search for the best match of a newly predicted spline segment to the last executed action to avoid boundary jumps under pipelined high-speed execution.
    Introduced in Sec. 4 Eq. 2 and Algorithm 2; ablation shows benefit, but no external independent validation.

pith-pipeline@v1.1.0-grok45 · 17267 in / 3645 out tokens · 41482 ms · 2026-07-13T01:28:09.564141+00:00 · methodology

0 comments
read the original abstract

In this work, we present B-spline Policy (BSP), an action representation designed for accelerating robot manipulation policies. Rather than predicting discrete-time action chunks, BSP parameterizes actions as continuous B-spline curves defined by a set of knots and control points. This representation yields smooth, time-continuous trajectories that can be temporally scaled and executed by low-level controllers at higher frequencies and speeds. We show that B-spline-parameterized actions can be seamlessly integrated into standard policy learning pipelines by directly predicting B-spline parameters. Experiments on simulated and real-world tasks demonstrate that BSP significantly reduces task completion time, achieving substantial improvements over baseline methods while maintaining strong success rates. More results: https://b-spline-policy.github.io

Figures

Figures reproduced from arXiv: 2607.09648 by Antonio Torralba, Chaoqi Liu, Haonan Chen, Haoyu Xiong, Xiaoshen Han, Yilun Du, Yuke Zhu.

Figure 1
Figure 1. Figure 1: B-spline policy. Rather than predicting a chunk of discrete-time actions, B-spline policy pa￾rameterizes actions as continuous B-spline[2]curves. This continuous representation enables temporal rescaling at execution time, allowing the policy to ex￾ecute faster and more smoothly. Robotic manipulation via visuomotor policy learning has made remarkable progress in re￾cent years [1, 2, 3, 4]. Yet, despite the… view at source ↗
Figure 2
Figure 2. Figure 2: Adaptive B-spline fitting for demonstra￾tion trajectories. Given a discrete-time demonstra￾tion trajectory, we convert it into a continuous B￾spline representation by adaptively selecting knots and estimating control points under a bounded recon￾struction error. Discrete Action Chunking. We focus on learn￾ing a robot policy π(a | o) from demonstra￾tion datasets. The policy maps proprioceptive and visual ob… view at source ↗
Figure 3
Figure 3. Figure 3: B-spline policy inference overview. At each inference step, the policy takes images and propriocep￾tion as the input and outputs future B-spline segment parameters. At the low-level control stage, we align each successive B-spline segment with the prior one. The predicted B-spline segment can be temporally scaled to adjust execution speed. Actions are then sampled at a higher frequency and sent to the robo… view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the real-world tasks. Task setups and corresponding object initialization distributions and workspace sizes for Cube Picking, Table Cleaning, and Speed Stacking. These tasks vary in precision requirements, horizon length, and object interactions, highlighting the challenges of fast manipulation due to tight precision requirements, long task horizons, and complex object interactions. Task Metric… view at source ↗
Figure 5
Figure 5. Figure 5: Real-world qualitative results. We visualize key frames from the rollouts of B-spline policy and a diffusion policy on the speed stacking task, with the robot gripper tip trajectories overlaid on the frame, illustrating that B-spline policy achieves faster and smoother task execution. 5.2 Real-world Results [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation study on inference-time seg￾ment alignment. Aligning adjacent segments im￾proves trajectory continuity, which becomes criti￾cal at high speeds, leading to higher success rates and more consistent realized speedups. Finding 4: Inference-time segment alignment is critical for BSP. Consistent with the smoother tra￾jectories observed in Finding 3, we further find that trajectory continuity across cons… 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

39 extracted references · 1 canonical work pages

  1. [1]

    Intelligence, K

    P. Intelligence, K. Black, N. Brown, J. Darpinian, K. Dhabalia, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusai, et al.π 0.5: A vision-language-action model with open-world generalization. arXiv preprint arXiv:2504.16054, 2025

    Pith/arXiv arXiv 2025

  2. [2]

    H. Wu, Y . Jing, C. Cheang, G. Chen, J. Xu, X. Li, M. Liu, H. Li, and T. Kong. Unleash- ing large-scale video generative pre-training for visual robot manipulation.arXiv preprint arXiv:2312.13139, 2023

    Pith/arXiv arXiv 2023

  3. [3]

    T. Z. Zhao, V . Kumar, S. Levine, and C. Finn. Learning fine-grained bimanual manipulation with low-cost hardware.arXiv preprint arXiv:2304.13705, 2023

    Pith/arXiv arXiv 2023

  4. [4]

    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.The International Journal of Robotics Research, 44(10-11):1684–1704, 2025

    2025

  5. [5]

    Q. Chen, J. Yu, M. Schwager, P. Abbeel, F. Shentu, and P. Wu. Sarm: Stage-aware reward modeling for long horizon robot manipulation, 2025. URLhttps://arxiv.org/abs/ 2509.25358

    Pith/arXiv arXiv 2025

  6. [6]

    Black, N

    K. Black, N. Brown, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusai, L. Groom, K. Haus- man, B. Ichter, S. Jakubczak, T. Jones, L. Ke, S. Levine, A. Li-Bell, M. Mothukuri, S. Nair, K. Pertsch, L. X. Shi, J. Tanner, Q. Vuong, A. Walling, H. Wang, and U. Zhilinsky.π 0: A vision-language-action flow model for general robot control, 2024. URLhttps://arxiv. o...

    Pith/arXiv arXiv 2024

  7. [7]

    Z. Fu, T. Z. Zhao, and C. Finn. Mobile aloha: Learning bimanual mobile manipulation with low-cost whole-body teleoperation. InConference on Robot Learning (CoRL), 2024

    2024

  8. [8]

    Y . Li, T. Huang, and D. G. Chetwynd. An approach for smooth trajectory planning of high- speed pick-and-place parallel robots using quintic b-splines.Mechanism and Machine Theory, 126:479–490, 2018

    2018

  9. [9]

    C. Wang, T. Xu, and M. Tomizuka. Continuous trajectory optimization via b-splines for multi- jointed robotic systems.arXiv preprint arXiv:2212.10500, 2022

    Pith/arXiv arXiv 2022

  10. [10]

    Wu.NURBS Curve Based Trajectory Design, pages 221–244

    G. Wu.NURBS Curve Based Trajectory Design, pages 221–244. Springer Nature Singapore, Singapore, 2025. ISBN 978-981-96-0443-2. doi:10.1007/978-981-96-0443-2 11. URL https://doi.org/10.1007/978-981-96-0443-2_11

    work page doi:10.1007/978-981-96-0443-2 2025

  11. [11]

    Natarajan, C

    R. Natarajan, C. Liu, H. Choset, and M. Likhachev. Implicit graph search for planning on graphs of convex sets, 2024. URLhttps://arxiv.org/abs/2410.08909

    Pith/arXiv arXiv 2024

  12. [12]

    Cohen, T

    E. Cohen, T. Lyche, and R. Riesenfeld. Discrete b-splines and subdivision techniques in computer-aided geometric design and computer graphics.Computer graphics and image pro- cessing, 14(2):87–111, 1980

    1980

  13. [13]

    R. H. Bartels, J. C. Beatty, and B. A. Barsky.An introduction to splines for use in computer graphics and geometric modeling. Morgan Kaufmann, 1995

    1995

  14. [14]

    De Boor and C

    C. De Boor and C. De Boor.A practical guide to splines, volume 27. springer New York, 1978

    1978

  15. [15]

    C. De Boor. On calculating with b-splines.Journal of Approximation theory, 6(1):50–62, 1972

    1972

  16. [16]

    P. Dierckx. An algorithm for smoothing, differentiation and integration of experimental data using spline functions.Journal of Computational and Applied Mathematics, 1(3):165–184, 1975. 9

    1975

  17. [17]

    C. De Boor. On uniform approximation by splines.J. Approx. Theory, 1(1):219–235, 1968

    1968

  18. [18]

    Ichnowski, Y

    J. Ichnowski, Y . Avigal, Y . Liu, and K. Goldberg. Gomp-fit: Grasp-optimized motion planning for fast inertial transport. In2022 international conference on robotics and automation (ICRA), pages 5255–5261. IEEE, 2022

    2022

  19. [19]

    K. Shaw, Y . Li, J. Yang, M. K. Srirama, R. Liu, H. Xiong, R. Mendonca, and D. Pathak. Bimanual dexterity for complex tasks. In8th Annual Conference on Robot Learning, 2024

    2024

  20. [20]

    Xiong, X

    H. Xiong, X. Xu, J. Wu, Y . Hou, J. Bohg, and S. Song. Vision in action: Learning active perception from human demonstrations.arXiv preprint arXiv:2506.15666, 2025

    Pith/arXiv arXiv 2025

  21. [21]

    M. Liu, Z. Zhu, X. Han, P. Hu, H. Lin, X. Li, J. Chen, J. Xu, Y . Yang, Y . Lin, et al. Ma- nipulation as in simulation: Enabling accurate geometry perception in robots.arXiv preprint arXiv:2509.02530, 2025

    Pith/arXiv arXiv 2025

  22. [22]

    Black, M

    K. Black, M. Galliker, and S. Levine. Real-time execution of action chunking flow policies. Advances in Neural Information Processing Systems, 38:33383–33407, 2026

    2026

  23. [23]

    L. Guo, Z. Xue, Z. Xu, and H. Xu. Demospeedup: Accelerating visuomotor policies via entropy-guided demonstration acceleration.arXiv preprint arXiv:2506.05064, 2025

    Pith/arXiv arXiv 2025

  24. [24]

    N. R. Arachchige, Z. Chen, W. Jung, W. C. Shin, R. Bansal, P. Barroso, Y . H. He, Y . C. Lin, B. Joffe, S. Kousik, et al. Sail: Faster-than-demonstration execution of imitation learning policies.arXiv preprint arXiv:2506.11948, 2025

    Pith/arXiv arXiv 2025

  25. [25]

    C. Liu, H. Chen, S. H. Høeg, S. Yao, Y . Li, K. Hauser, and Y . Du. Flexible multitask learning with factorized diffusion policy, 2025. URLhttps://arxiv.org/abs/2512.21898

    Pith/arXiv arXiv 2025

  26. [26]

    X. Han, M. Liu, Y . Chen, J. Yu, X. Lyu, Y . Tian, B. Wang, W. Zhang, and J. Pang. Re 3sim: Generating high-fidelity simulation data via 3d-photorealistic real-to-sim for robotic manipu- lation.arXiv preprint arXiv:2502.08645, 2025

    Pith/arXiv arXiv 2025

  27. [27]

    Block, A

    A. Block, A. Jadbabaie, D. Pfrommer, M. Simchowitz, and R. Tedrake. Provable guarantees for generative behavior cloning: Bridging low-level stability and high-level behavior. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors,Advances in Neural Information Processing Systems, volume 36, pages 48534–48547. Curran Associates, Inc.,

  28. [28]

    URLhttps://proceedings.neurips.cc/paper_files/paper/2023/ file/97c903fbf21a7d863af2015d8803ca8f-Paper-Conference.pdf

    2023

  29. [29]

    Malik, T

    A. Malik, T. Henderson, and R. J. Prazenica. Trajectory generation for a multibody robotic system using the product of exponentials formulation. InAIAA Scitech 2021 Forum, page 2016, 2021

    2021

  30. [30]

    Grotjahn and B

    M. Grotjahn and B. Heimann. Model-based feedforward control in industrial robotics.The International Journal of Robotics Research, 21(1):45–60, 2002

    2002

  31. [31]

    C. An, C. Atkeson, J. Griffiths, and J. Hollerbach. Experimental evaluation of feedforward and computed torque control. InProceedings. 1987 IEEE International Conference on Robotics and Automation, volume 4, pages 165–168. IEEE, 1987

    1987

  32. [32]

    H. Zhou, W. Liao, X. Huang, Y . Tang, F. Otto, X. Jia, X. Jiang, S. Hilber, G. Li, Q. Wang, et al. Beast: Efficient tokenization of b-splines encoded action sequences for imitation learning. arXiv preprint arXiv:2506.06072, 2025

    arXiv 2025

  33. [33]

    A. J. Ijspeert, J. Nakanishi, H. Hoffmann, P. Pastor, and S. Schaal. Dynamical movement primitives: learning attractor models for motor behaviors.Neural computation, 25(2):328– 373, 2013. 10

    2013

  34. [34]

    T. T. Zhang, D. Pfrommer, C. Pan, N. Matni, and M. Simchowitz. Action chunking and ex- ploratory data collection yield exponential improvements in behavior cloning for continuous control, 2025. URLhttps://arxiv.org/abs/2507.09061

    arXiv 2025

  35. [35]

    Dierckx.Curve and surface fitting with splines

    P. Dierckx.Curve and surface fitting with splines. Oxford University Press, 1995

    1995

  36. [36]

    P. Dierckx. Algorithms for smoothing data with periodic and parametric splines.Computer Graphics and Image Processing, 20(2):171–184, 1982

    1982

  37. [37]

    Florence, C

    P. Florence, C. Lynch, A. Zeng, O. A. Ramirez, A. Wahid, L. Downs, A. Wong, J. Lee, I. Mor- datch, and J. Tompson. Implicit behavioral cloning. InConference on robot learning, pages 158–168. PMLR, 2022

    2022

  38. [38]

    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. InarXiv preprint arXiv:2108.03298, 2021

    Pith/arXiv arXiv 2021

  39. [39]

    Nasiriany, A

    S. Nasiriany, A. Maddukuri, L. Zhang, A. Parikh, A. Lo, A. Joshi, A. Mandlekar, and Y . Zhu. Robocasa: Large-scale simulation of everyday tasks for generalist robots.arXiv preprint arXiv:2406.02523, 2024. 11 A Appendix A.1 Implement Details B-spline action representation.We represent action trajectories using cubic B-splines with degree k= 3. In simulatio...

    Pith/arXiv arXiv 2024