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

TL;DR

SAIL introduces a comprehensive system that enables imitation learning policies to execute tasks at speeds up to four times faster than the demonstration data, enhancing robotic throughput in real-world applications.

Contribution

The paper formalizes the problem of faster-than-demonstration execution and presents SAIL, a full-stack system with four components to achieve high-speed visuomotor policy execution.

Findings

Up to 4x speedup in simulation

Up to 3.2x speedup in real robots

Effective across diverse tasks and platforms

Abstract

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