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

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Arachchige, Nadun Ranawaka, Chen, Zhenyang, Jung, Wonsuhk, Shin, Woo Chul, Bansal, Rohan, Barroso, Pierre, He, Yu Hang, Lin, Yingyang Celine, Joffe, Benjamin, Kousik, Shreyas, Xu, Danfei
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911142247399424
author Arachchige, Nadun Ranawaka
Chen, Zhenyang
Jung, Wonsuhk
Shin, Woo Chul
Bansal, Rohan
Barroso, Pierre
He, Yu Hang
Lin, Yingyang Celine
Joffe, Benjamin
Kousik, Shreyas
Xu, Danfei
author_facet Arachchige, Nadun Ranawaka
Chen, Zhenyang
Jung, Wonsuhk
Shin, Woo Chul
Bansal, Rohan
Barroso, Pierre
He, Yu Hang
Lin, Yingyang Celine
Joffe, Benjamin
Kousik, Shreyas
Xu, Danfei
contents 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
format Preprint
id arxiv_https___arxiv_org_abs_2506_11948
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SAIL: Faster-than-Demonstration Execution of Imitation Learning Policies
Arachchige, Nadun Ranawaka
Chen, Zhenyang
Jung, Wonsuhk
Shin, Woo Chul
Bansal, Rohan
Barroso, Pierre
He, Yu Hang
Lin, Yingyang Celine
Joffe, Benjamin
Kousik, Shreyas
Xu, Danfei
Robotics
Artificial Intelligence
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
title SAIL: Faster-than-Demonstration Execution of Imitation Learning Policies
topic Robotics
Artificial Intelligence
url https://arxiv.org/abs/2506.11948