FODMP: Fast One-Step Diffusion of Movement Primitives Generation for Time-Dependent Robot Actions
Authors: Xirui Shi, Arya Ebrahimi, Yi Hu, Jun Jin
Organizations: Department of Electrical and Computer Engineering, University of Alberta, Edmonton, Alberta, Canada. · Jun Jin is also with the Alberta Machine Intelligence Institute(Amii), Edmonton, Alberta, Canada.
Diffusion models are increasingly used for robot learning, but current designs face a clear trade-off. Action-chunking diffusion policies like ManiCM are fast to run, yet they only predict short segments of motion. This makes them reactive, but unable to capture time-dependent motion primitives, such as following a spring-damper-like behavior with built-in dynamic profiles of acceleration and deceleration. Recently, Movement Primitive Diffusion (MPD) partially addresses this limitation by parameterizing full trajectories using Probabilistic Dynamic Movement Primitives (ProDMPs), thereby enabling the generation of temporally structured motions. Nevertheless, MPD integrates the motion decoder directly into a multi-step diffusion process, resulting in prohibitively high inference latency that limits its applicability in real-time control settings. We propose FODMP (Fast One-step Diffusion of Movement Primitives), a new framework that distills diffusion models into the ProDMPs trajectory parameter space and generates motion using a single-step decoder. FODMP retains the temporal structure of movement primitives while eliminating the inference bottleneck through single-step consistency distillation. This enables robots to execute time-dependent primitives at high inference speed, suitable for closed-loop vision-based control. On standard manipulation benchmarks (MetaWorld, ManiSkill), FODMP runs up to 10 times faster than MPD and 7 times faster than action-chunking diffusion policies, while matching or exceeding their success rates. Beyond speed, by generating fast acceleration-deceleration motion primitives, FODMP allows the robot to intercept and securely catch a fast-flying ball, whereas action-chunking diffusion policy and MPD respond too slowly for real-time interception.
Figures & tables
Fig. 1: Comparison of Policy Designs. (a) Action chunking–based diffusion policies predict discrete sequences of actions without explicitly modeling time-dependent motion, resulting in temporally unstructured behavior. (b) MPD introduces a motion decoder into the diffusion process to recover time-dependent trajectories; however, motion parameters are still inferred through multi-step denoising, leading to high inference latency. (c) Our method directly predicts movement primitive parameters via one-step diffusion, followed by a single decoder pass to generate time-dependent motion. This design enables low-latency inference while explicitly modeling continuous motion evolution over time.
Fig. 2: Method Overview. Our approach conditions a Transformer-based one-step diffusion model on multi-modal observations to directly predict ProDMPs parameters. The predicted parameters are decoded by a ProDMPs ODE solver, enabling fast generation of smooth, structured robot trajectories across diverse robot tasks.
Fig. 3: Consistency Distillation Pipeline. Given observations, a multi-step teacher performs k-step denoising to obtain cleaner parameters, which supervise a one-step student model. The student is trained using a consistency objective against a target model, whose parameters are updated via EMA from the student.
Methods
Easy
Medium
Hard
Average
Success Rate (%)
DP
99.3 ± 0.1
41.0 ± 3.2
10.1 ± 1.4
50.1
ManiCM
79.2 ± 0.4
18.9 ± 5.1
5.2 ± 0.2
34.4
MPD
98.9 ± 0.3
64.8 ± 2.6
28.6 ± 2.9
64.1
FODMP
99.2 ± 0.1
86.3 ± 1.2
49.0 ± 2.3
78.2
Inference Time (ms)
TABLE I: Success rate and inference time across all tasks based on our models and baselines. The best results for each category are in bold font and the second best ones are underlined for an easier comparison.
Fig. 4: Simulation Environments. We evaluate our method on a diverse set of manipulation tasks from MetaWorld and ManiSkill, grouped by difficulty into easy, medium, and hard categories.
Fig. 5: RealWorld Experiments Setups. (a) Push-T task setup, where a Franka arm pushes a T-shaped object using visual observations from two RealSense cameras. (b) Ball-catching task setup, where the robot intercepts a human-thrown ball using depth observations.
Fig. 6: RealWorld Experiments Results. Success rates of different methods on (a) the Push-T task and (b) the Ball Catching task.
Fig. 7: Ball Catching Task Performance Comparison. Snapshots over time illustrate that baseline methods fail to adapt to the fast-moving ball, whereas FODMP produces temporally consistent motions that successfully catch and stabilize the ball.
Fig. 8: Data Effiency Comparison. (a) Ball Catching task and (b) Push-T task. Success rates of different methods under varying numbers of training demonstrations.
Fig. 9: Motion analysis. The temporal evolution of (a) end-effector velocities and (b) positions along Cartesian axes. The smooth velocity transitions and continuous spatial path indicate temporally consistent motion generation
Robot motion generation creates feasible, task-directed trajectories for robots to move, manipulate objects, and interact with people and environments.
Generative flow and diffusion models provide the continuous, multimodal action distributions needed for high-precision robotic policies. However, their reliance on iterative sampling introduces severe inference latency, degrading control frequency and harming performance in time-sensitive manipulation. To address this problem, we propose the One-Step Flow Policy (OFP), a from-scratch self-distillation framework for high-fidelity, single-step action generation without a pre-trained teacher. OFP unifies a self-consistency loss to enforce coherent transport across time intervals, and a self-guided regularization to sharpen predictions toward high-density expert modes. In addition, a warm-start mechanism leverages temporal action correlations to minimize the generative transport distance. Evaluations across 56 diverse simulated manipulation tasks demonstrate that a one-step OFP achieves state-of-the-art results, outperforming 100-step diffusion and flow policies while accelerating action generation by over 100×. We further integrate OFP into the π0.5 model on RoboTwin 2.0, where one-step OFP surpasses the original 10-step policy. These results establish OFP as a practical, scalable solution for highly accurate and low-latency robot control.
Shaolong Li, Lichao Sun, Yongchao Chen
CSE Department, University of Michigan · Lehigh University · College of AI, Tsinghua University
Diffusion models excel at generating diverse, multimodal trajectories for robotic control, yet their iterative denoising process introduces latency that can constrain real-time closed-loop execution. To address this problem, we propose a Dynamic Neural Koopman (DNK) distillation framework, which distills multistep diffusion inference into a single forward pass conditioned on observations and sampled noise. Specifically, we introduce a Factorized Dynamic Koopman (FDK) layer that models the denoising process through a latent linear transition. The noisy trajectory input is lifted into a latent space, where the FDK layer parameterizes a factorized linear operator with condition-dependent eigenvalues that adapt the noisy-to-denoised transition to the current context. We evaluate DNK across robot-control benchmarks that span locomotion, state-based, and image-based manipulation, comparing against accelerated generative-policy baselines. The results demonstrate that DNK achieves competitive or improved closed-loop performance relative to accelerated generative-policy baselines, while substantially reducing inference latency relative to the strongest high-performing one-step baselines. On long-horizon manipulation tasks, DNK achieves competitive performance with substantially fewer inference parameters than the one-step policy baselines. Hardware experiments on a physical Kinova manipulator further demonstrate millisecond-scale inference and reduced task completion time. A project page is available at https://fdkoopman.github.io/.
Lei Zheng, Peiqi Yu, Zengqi Peng +2
Department of Electrical and Computer Engineering, National University of Singapore, Singapore · Department of Electrical and Computer Engineering, Carnegie Mellon University, USA · Robotics and Autonomous Systems Thrust, The Hong Kong University of Science and Technology, Hong Kong SAR, China
Diffusion-based visuomotor policies operating directly in raw action spaces conflate scene comprehension with trajectory generation within a single denoising process. The resulting velocity field must simultaneously encode scene information and generate precise trajectories, increasing learning complexity and limiting performance on tasks demanding precise temporal coordination across multiple arms. To simplify this joint learning problem, we introduce Latent Diffusion Policy (LDP), a two-stage framework performing flow matching in a deliberately shaped latent space. By absorbing scene understanding into an observation-conditioned CVAE encoder, LDP concentrates the conditional distribution of each observation. Consequently, the flow model avoids implicitly resolving scene-dependent structures; instead, it generates within a pre-concentrated distribution featuring a smoother velocity field, simplifying learning from limited demonstrations. Furthermore, to capture temporal dependencies among latent tokens, LDP trains with per-token diffusion forcing and employs staircase inference sampling to resolve the resulting distributional mismatch. We also propose reconstruction FID (rFID) as a lightweight proxy predicting downstream task success solely from latent space statistics. On coordination-intensive tasks from RoboTwin 2.0, LDP outperforms DP3 by a substantial margin and transfers effectively to real-world bimanual deployments.