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

2025-06-13

AI Summary

Existing imitation learning strategies are constrained by demonstration speed, limiting task throughput and hindering deployment in industrial automation. This work formally introduces the “super-speed execution” problem—executing tasks faster than demonstrated—and identifies its core challenges as robot dynamics constraints and state-action distributional shift. To address these, we propose a four-module synergistic framework: (1) consistency-preserving action inference, (2) controller-agnostic motion tracking, (3) dynamic complexity-aware adaptive speed modulation, and (4) real-time latency-compensated action scheduling. The method integrates smooth action-space inference, high-fidelity motion target tracking, variable-speed regulation, and robust scheduling. Evaluated across 12 tasks, it achieves up to 4× demonstration speed in simulation and 3.2× on physical robots, while maintaining stable, reliable execution.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Search and Optimization: Sampling/Simulation-based Search
• Intelligent Robots: Manipulation

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

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

Problem

• Enabling faster-than-demonstration execution of visuomotor policies
• Addressing robot dynamics and state-action distribution shifts
• Increasing task throughput for industrial automation applications

Innovation

• Consistency-preserving action inference for smooth high-speed motion
• High-fidelity tracking of controller-invariant motion targets
• Adaptive speed modulation based on motion complexity