Robots are increasingly learning new skills by observing humans. Many real-world human tasks, from folding laundry to handling food, are too subtle to be efficiently programmed step-by-step.
In imitation learning, a human demonstrates a task and the robot learns to copy what it sees through a camera or sensor. Although this approach is at the cutting edge of robotics research, it is limited by significant constraints. That is, a robot can only move as fast as the human who taught it.
Now, researchers at Georgia Tech have developed a tool to break that speed barrier. This system allows robots to perform complex tasks significantly faster than human demonstration while maintaining precision, control, and safety.
The team is tackling a central challenge in modern robotics: how to combine the flexibility of learning from humans with the speed and reliability needed for real-world deployment. This technology could lead to widespread adoption of imitative learning in industrial and domestic applications, and even enable robots to perform human-like tasks better than ever before.
“What we are trying to create, and what industry is trying to create, is a general-purpose robot that can do all the tasks that humans can do,” said Shreyas Kusik, assistant professor in the George W. Woodruff School of Mechanical Engineering and co-lead author of the study. “Speed is critical to making it work outside the lab.”
The new tool, SAIL (Speed Adaptation for Imitation Learning), is the result of a cross-campus, interdisciplinary collaboration that brings together expertise in mechanical engineering, robotics systems, and machine learning. The research team also includes Kusik. Benjamin Joffe, Senior Research Scientist, Georgia Tech Research Institute; Danfei Xu, assistant professor in the Department of Interactive Computing, and graduate students and researchers from multiple labs participated.
speed without sacrifice
It is difficult to teach a robot to move faster than a human can demonstrate. Robots can behave differently at higher speeds, and small changes in the environment can cause errors.
“The challenge is that the robots are limited to the data they were trained on, and changes in the environment can cause them to fail,” Kusik says.
SAIL addresses this challenge through a modular approach where separate components work together to accelerate beyond training data. The system maintains fast and smooth movements, accurately tracks movements, dynamically adjusts speed based on task complexity, and schedules actions to account for hardware delays. This combination allows robots to move rapidly while remaining stable, coordinated, and precise.
“One of the gaps that we saw was that our academic robotic systems could do great things, but they weren’t fast or robust enough for practical use,” Joffe said. “We wanted to carefully study that gap and design a system that addresses it end-to-end.”
He added, “The goal is not only to make robots faster, but also to make them smart enough to recognize when speed is useful and when it can cause mistakes.”
The team evaluated SAIL’s performance across 12 tasks, both in simulation and on two physical robot platforms. Tasks included stacking cups, folding cloth, plating fruit, packaging food, and wiping down whiteboards. In most cases, SAIL-enabled robots completed tasks three to four times faster than standard imitation learning systems, without compromising accuracy.
One exception was the whiteboard wiping task, which was difficult to perform quickly due to maintaining contact.
“It’s important to understand where speed helps and where it hurts,” Kusik says. “Sometimes slowing down is the right decision.”
Although SAIL does not by itself make robots universally adaptable, it is an important step toward robotic systems that can learn from humans without being limited by their own pace.
By showing how to safely and systematically accelerate learned robot movements, SAIL brings imitation learning closer to real-world use, where speed, accuracy, and reliability are all important.
Citation: Ranawaka Arachchige, et al. Al. “SAIL: Faster-than-Demonstration Execution of Imitation Learning Policies”, Conference on Robotic Learning (CoRL), 2025.
DOI: https://doi.org/10.48550/arXiv.2506.11948
Funding: The authors would like to thank the State of Georgia and the Georgia Tech Agricultural Technology Research Program for supporting the research described in this paper.
