mlsanityData to 14 Oct 2026

SAIL: Faster-than-Demonstration Execution of Imitation Learning Policies

arXiv 2506.11948 ↗First posted 13 Jun 2025cs.RO cs.AIcited by 18

On the map: Offline RL and imitation learning from data → Robot manipulation imitation learning

Abstract (arXiv)

Offline Imitation Learning (IL) methods such as Behavior Cloning are effective at acquiring complex robotic manipulation skills. However, existing IL-trained policies are confined to executing the task at the same speed as shown in demonstration data. This limits the task throughput of a robotic system, a critical requirement for applications such as industrial automation. In this paper, we introduce and formalize the novel problem of enabling faster-than-demonstration execution of visuomotor policies and identify fundamental challenges in robot dynamics and state-action distribution shifts. We instantiate the key insights as SAIL (Speed Adaptation for Imitation Learning), a full-stack system integrating four tightly-connected components: (1) a consistency-preserving action inference algorithm for smooth motion at high speed, (2) high-fidelity tracking of controller-invariant motion targets, (3) adaptive speed modulation that dynamically adjusts execution speed based on motion complexity, and (4) action scheduling to handle real-world system latencies. Experiments on 12 tasks across simulation and two real, distinct robot platforms show that SAIL achieves up to a 4x speedup over demonstration speed in simulation and up to 3.2x speedup in the real world. Additional detail is available at https://nadunranawaka1.github.io/sail-policy

Key facts. Cited by 18 papers in the map (data through 14 Oct 2026). Placed in Robot manipulation imitation learning, one of 13 directions in Offline RL and imitation learning from data. Builds on 25 of its 47 references that are in the map. First posted on arXiv 13 Jun 2025.

Builds on

25 of 47 references.

Closest ancestors · 11

Same direction, newest first

  1. 2024What's the Move? Hybrid Imitation Learning via Salient Points6 Dec
  2. 2024Open-TeleVision: Teleoperation with Immersive Active Visual Feedback1 Jul
  3. 2024Streaming Diffusion Policy: Fast Policy Synthesis with Variable Noise Diffusion Models7 Jun
  4. 2024Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation4 Jan
  5. 2023MimicGen: A Data Generation System for Scalable Robot Learning using Human Demonstrations26 Oct
  6. 2023Waypoint-Based Imitation Learning for Robotic Manipulation26 Jul
  7. 2023Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware23 Apr
  8. 2022VIOLA: Imitation Learning for Vision-Based Manipulation with Object Proposal Priors20 Oct
  9. 2022Imitation Learning for Nonprehensile Manipulation through Self-Supervised Learning Considering Motion Speed21 Jun
  10. 2021Imitation learning for variable speed motion generation over multiple actions11 Mar
  11. 2021Imitation Learning for Variable Speed Contact Motion for Operation up to Control Bandwidth20 Feb

Other references · 14

  1. 2022Classifier-Free Diffusion Guidancecited by 5,847
  2. 2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusioncited by 2,838
  3. 2021What Matters in Learning from Offline Human Demonstrations for Robot Manipulationcited by 725
  4. 2024RDT-1B: a Diffusion Foundation Model for Bimanual Manipulationcited by 691
  5. 2021Offline Reinforcement Learning as One Big Sequence Modeling Problemcited by 644
  6. 2020robosuite: A Modular Simulation Framework and Benchmark for Robot Learningcited by 511
  7. 2024Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robotscited by 394
  8. 2019Extrapolating Beyond Suboptimal Demonstrations via Inverse Reinforcement Learning from Observationscited by 260
  9. 2019Imitation Learning from Imperfect Demonstrationcited by 114
  10. 2024OPEN TEACH: A Versatile Teleoperation System for Robotic Manipulationcited by 71
  11. 2020Neural Dynamic Policies for End-to-End Sensorimotor Learningcited by 58
  12. 2021Dynamic Movement Primitives in Robotics: A Tutorial Surveycited by 47
2 more
  1. 2018Riemannian Motion Policiescited by 46
  2. 2021Hierarchical Neural Dynamic Policiescited by 17

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 4; 2026: 14.

18 papers.

Sort:

By area:

  1. 2025ActiveUMI: Robotic Manipulation with Active Perception from Robot-Free Human Demonstrationscited by 19
  2. 2026Learning Native Continuation for Action Chunking Flow Policiescited by 19
  3. 2026Tune to Learn: How Controller Gains Shape Robot Policy Learningcited by 8
  4. 2025SpeedAug: Policy Acceleration via Tempo-Enriched Policy and RL Fine-Tuningcited by 4
  5. 2025ESPADA: Execution Speedup via Semantics Aware Demonstration Data Downsampling for Imitation Learningcited by 4
  6. 2026Speedup Patch: Learning a Plug-and-Play Policy to Accelerate Embodied Manipulationcited by 3
  7. 2026TempoVLA: Learning Speed-Controllable Vision-Language-Action Policiescited by 3
  8. 2026From Inference Efficiency to Embodied Efficiency: Revisiting Efficiency Metrics for Vision-Language-Action Modelscited by 2
  9. 2026Flip Stunts on Bicycle Robots using Iterative Motion Imitationcited by 2
  10. 2025When a Robot is More Capable than a Human: Learning from Constrained Demonstratorscited by 1
  11. 2026Habilis-β: A Fast-Motion and Long-Lasting On-Device Vision-Language-Action Modelcited by 1
  12. 2026Refinement of Accelerated Demonstrations via Incremental Iterative Reference Learning Control for Fast Contact-Rich Imitation Learningcited by 1
6 more
  1. 2026OmniTacTune: Policy-Agnostic Real-World RL for Tactile Residual Adaptation of Visual Policiescited by 1
  2. 2026Device-Conditioned Neural Architecture Search for Efficient Robotic Manipulationcited by 0
  3. 2026SkiP: When to Skip and When to Refine for Efficient Robot Manipulationcited by 0
  4. 2026LAGO Policy: Latency-Aware Asynchronous Diffusion Policies with Goal-Directed Collision-Free Planning for Smooth Manipulationcited by 0
  5. 2026Does Imitation Learning Preserve Temporal Robustness in Dexterous Manipulation? An Expert-Learner Comparison Across Task Execution Speedscited by 0
  6. 2026EgoSpeedUp: Transferring Human Manipulation Tempo to Robot Policiescited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026SAIL: Test-Time Scaling for In-Context Imitation Learning with VLM
  2. 2026AdaTempo: Learning Shared Relative Tempo from Demonstrations for Faster Robot Manipulation
  3. 2026SpeedTuning: Speeding Up Policy Execution with Lightweight Reinforcement Learning
  4. 2020Learning Sparse Rewarded Tasks from Sub-Optimal Demonstrations
  5. 2025DemoSpeedup: Accelerating Visuomotor Policies via Entropy-Guided Demonstration Acceleration
  6. 2025Robust Offline Imitation Learning Through State-level Trajectory Stitching
  7. 2020Concurrent Training Improves the Performance of Behavioral Cloning from Observation
  8. 2025Learning a Thousand Tasks in a Day