Smooth Action Chunking Via Movement-Primitive Representations for Imitation Learning

Shota Nomura; Takayuki Osa; Yasuhiro Kato; Kazuki Ota; Naoki Shitanda; Tatsuya Harada

cs.LGOct 29, 2026

Smooth Action Chunking Via Movement-Primitive Representations for Imitation Learning

Authors: Shota Nomura, Takayuki Osa, Yasuhiro Kato, Kazuki Ota, Naoki Shitanda, Tatsuya Harada

Organizations: The University of Tokyo, Tokyo, Japan. · RIKEN Center for Advanced Intelligence Project, Tokyo, Japan.

Abstract

Imitation learning has become a widely adopted approach for vision-based robot policies, owing to its conceptual simplicity, scalability, and ability to leverage large-scale datasets. As these policies have grown more capable, action chunking has become their standard interface. At each decision step the policy outputs several future actions at once. Backbones from diffusion policies to large vision-language-action models all share it. Whatever backbone produces the chunk, its temporal structure is learned only implicitly. The commanded trajectory is therefore jerky both within a chunk and between successive chunks. On real hardware this appears as visible vibration and as repeated saturation of the follower's velocity limit. We address this issue by incorporating a movement-primitive-based trajectory representation into action chunking. Specifically, a trajectory segment is represented as a linear combination of smooth basis functions, yielding smooth trajectories by construction while leaving the policy architecture, training objective, and inference procedure unchanged. The same representation also yields a closed-form, inference-time correction for the second source of jerk. We condition the predicted coefficients so that the new chunk passes through a via-point at the previous chunk's last commanded action. This correction needs no retraining and, unlike asynchronous Real-Time Chunking, applies to policies queried in one shot. We instantiate it on a one-shot (ACT) and an iterative (DiT) backbone and evaluate in simulation and on a real bimanual ALOHA platform. The representation lowers within-chunk jerk by an order of magnitude on the one-shot backbone and by 2.4 times on the iterative one, with task success statistically non-inferior. The conditioning then removes 73-84% of the boundary discontinuity, at negligible per-step cost.

Figures & tables

Fig. 1 : Overview of this study. Top left: predicting the raw action chunk leaves the commanded trajectory jerky within a chunk and discontinuous between chunks. Top right: predicting it as a weighted sum of smooth basis functions is smooth by construction. Bottom: the backbone predicts movement-primitive coefficients $\hat{\bm{w}}$ . At inference we condition them on the previous chunk’s last point to obtain the corrected coefficients $\hat{\bm{w}}^{\prime}$ , then decode these into the action chunk.

Panel of Figure 2 — Fig. 2 : Evaluation tasks. (a, b) Simulation. (c–e) Real ALOHA.

(a) Success rate (%) ( $\uparrow$ )
Backbone	Task	Baseline	+MP	+Cond
ACT	Transfer Cube	$96.0\pm 5.3$	$97.7\pm 1.2$	$99.0\pm 1.0$
ACT	Insertion	$47.7\pm 15.8$	$55.7\pm 13.4$	$54.3\pm 6.4$
DiT	Transfer Cube	$85.7\pm 0.6$	$77.3\pm 3.5$	$78.0\pm 6.9$
DiT	Insertion	$57.0\pm 10.0$	$55.3\pm 5.5$	$63.0\pm 8.9$
(b) Within-chunk commanded jerk ( $\downarrow$ )

TABLE I : Simulation results. +Cond is +MP with inter-chunk conditioning. Arrows indicate the better direction, and bold marks the lowest jerk in each row of (b) and (c).

(a) Success rate (%) ( $\uparrow$ )
		Raw action			MP
Backbone	Task	Baseline	+WS	+RTC	+MP	+WS	+Cond
ACT	Pick-and-Pass	$80.0\pm 10.0$	$83.3\pm 20.8$	–	$73.3\pm 23.1$	$76.7\pm 20.8$	$83.3\pm 15.3$
	Fold Cloth	$96.7\pm 5.8$	$96.7\pm 5.8$	–	$93.3\pm 5.8$	$100.0\pm 0.0$	$100.0\pm 0.0$
	Peg Insertion	$80.0\pm 17.3$	$86.7\pm 15.3$	–	$80.0\pm 10.0$	$76.7\pm 5.8$	$70.0\pm 10.0$
DiT	Pick-and-Pass	$86.7\pm 15.3$	$96.7\pm 5.8$	$83.3\pm 20.8$	$90.0\pm 10.0$	$96.7\pm 5.8$	$96.7\pm 5.8$

TABLE II : Real-hardware results. Columns are grouped by the representation they run on, and columns within a group share the same trained weights. Arrows indicate the better direction, and bold marks the lowest value in each row of (b)–(e). RTC applies to DiT only.

Fig. 3 : Commanded trajectory around a chunk boundary on Fold Cloth. (a) Command of one representative joint on ACT with no connection, blending, and conditioning. (b) Deviation of the command from the freshly predicted chunk, averaged over arm joints, against time since it arrived, on ACT and DiT. Shading marks the integral.

$\Sigma_{y}^{*}$	$m$	Boundary ( $\downarrow$ )	Executed $1$ – $5\,$ Hz ( $\downarrow$ )	Success (%) ( $\uparrow$ )
$10^{-6}$	1	$75\pm 3$	$6.69\pm 0.17$	$83\pm 6$
$10^{-3}$	1	$73\pm 2$	$6.50\pm 0.10$	$83\pm 15$
$10^{-2}$	1	$81\pm 4$	$5.42\pm 0.18$	$87\pm 12$
$10^{-1}$ $\dagger$	1	$121\pm 5$	$\mathbf{4.76\pm 0.13}$	$83\pm 15$
$10^{0}$	1	$398\pm 47$	$5.31\pm 0.36$	$83\pm 29$
$10^{+1}$	1	$740\pm 80$	$6.40\pm 0.51$	$80\pm 17$

TABLE III : Conditioning ablation on Pick-and-Pass with ACT. All rows use the same +MP model.

\dagger

marks the selected operating point.

Robot Imitation Learning

Robot imitation learning trains robots to perform tasks by reproducing and adapting behaviors shown in demonstrations.

See all 691 papers

Apr 7, 2026

HiPolicy: Hierarchical Multi-Frequency Action Chunking for Policy Learning

Robotic imitation learning faces a fundamental trade-off between modeling long-horizon dependencies and enabling fine-grained closed-loop control. Existing fixed-frequency action chunking approaches struggle to achieve both. Building on this insight, we propose HiPolicy, a hierarchical multi-frequency action chunking framework that jointly predicts action sequences at different frequencies to capture both coarse high-level plans and precise reactive motions. We extract and fuse hierarchical features from history observations aligned to each frequency for multi-frequency chunk generation, and introduce an entropy-guided execution mechanism that adaptively balances long-horizon planning with fine-grained control based on action uncertainty. Experiments on diverse simulated benchmarks and real-world manipulation tasks show that HiPolicy can be seamlessly integrated into existing 2D and 3D generative policies, delivering consistent improvements in performance while significantly enhancing execution efficiency.

Jiyao Zhang, Zimu Han, Junhan Wang +7

CFCS, School of CS, PKU · National Key Laboratory for Multimedia Information Processing, School of CS, PKU · XJTU +3

Policy Learning Action Chunking Robot Imitation Learning Long-Horizon Robotic Manipulation Hierarchical RL

Ask PDF

Aug 3, 2026

Why Does Action Chunking Improve Behavioral Cloning Performance in Robotic Control?

Action chunking---predicting and executing multiple actions instead of a single action---has proven to be a critical component for learning effective robotic control policies. However, our precise understanding of why action chunking improves performance has remained limited. In this work we seek to close this gap. Through rigorous experimental evaluations in both simulated and real-world settings, we show that existing hypotheses for the success of action chunking---temporal consistency, horizon reduction, and representation learning---fail to explain the success of action chunking. Instead, we find that action chunking benefits from greater non-Markovian expressivity and reduced compounding error compared to Markovian policies, but, in many settings of interest, these effects can be fully captured by delayed policies, which at each step predict a single action based on the observation

k

steps in the past. We then show that there exists an additional benefit of action chunking that we refer to as implicit ensembling. In particular, by learning a diversity of temporal relationships (that is,

a_t | o_t, a_t | o_{t-1}, \ldots

), action-chunked policies exhibit behavior matching that of a model ensemble, increasing their robustness and generalization ability over policies that only learn a single temporal relationship. Building on these insights, we show that in simulated and real-world robotic control settings, we can match the performance of action chunking without action chunking---by deploying an action chunking policy as an ensemble of policies with randomized delays. Furthermore, we propose a policy class that amplifies the benefits of action chunking by explicitly instantiating an ensemble, and which we show significantly improves over the performance of action chunking in many domains.

Filippo Lazzati, Kyle Stachowicz, William Chen +3

Politecnico di Milano · UC Berkeley

Ensemble Learning Robot Imitation Learning Action Chunking Robot Policy Generalization

Ask PDF

Jul 31, 2026

TRACT: Temporally Routed Action Chunks with Chronological Phase Authority for Contact-Rich Manipulation

Action chunking shortens the effective decision horizon of robot imitation learning by predicting multiple future actions, while conventional phase conditioning describes the current control instant. When a predicted horizon crosses a procedural boundary, assigning the current phase to the entire chunk creates a structural temporal mismatch. We present TRACT, which factorizes phase-structured action chunking into an accepted current phase and a single CURRENT-to-NEXT boundary inside the future horizon. A task-local graph constrains chronological phase authority, and a cumulative boundary distribution monotonically routes future queries through phase-specific query and action paths. For contact execution, a causal response-deficit integrator compares policy intent with ACK-eligible subsequent motion, accumulates arm compensation when directional response is suppressed, and decays after confirmed recovery. Across six real-robot variants with ten trials each, full TRACT achieves 10/10 full-sequence success, 99.00 [88.75, 100.00]% median [min, max] wipe completion, zero observed phase ambiguity, and zero stalls. Under the current complete method package and evaluation setting, the routed representation obtains better observed task results than the flat package (6/10 vs. 3/10 success; 77.08% vs. 8.03% median wipe completion). Chronological authority reduces observed phase ambiguity from 8/10 to 0/10, and response integration reduces stalls from 4/10 to 0/10. The package comparison does not isolate routing from other generator-package differences.

Jiahao Liu, Kento Kawaharazuka, Tasuku Makabe +1

Department of Mechano-Informatics, Graduate School of Information Science and Technology, The University of Tokyo, Tokyo, Japan

Robotic Control Robot Skill Learning Contact-Rich Robotic Manipulation Robot Imitation Learning Action Chunking

Ask PDF

Abstract

Robot Imitation Learning

Explore similar work

HiPolicy: Hierarchical Multi-Frequency Action Chunking for Policy Learning

Why Does Action Chunking Improve Behavioral Cloning Performance in Robotic Control?

TRACT: Temporally Routed Action Chunks with Chronological Phase Authority for Contact-Rich Manipulation