Presentation Type

Oral/Paper Presentation

Abstract

Teaching robots through reinforcement learning (RL) has made great progress, but real-world training is still difficult. Robots need lots of practice to learn, rewards can be hard to define, and resetting the environment after each attempt is often a challenge. The Sample-Efficient Robotic Reinforcement Learning (SERL) framework helps solve these issues by offering a ready-to-use, open-source software package that makes RL more practical for real-world robotics.

This project explores SERL and how it improves robotic RL by making learning faster and more efficient. SERL includes smarter ways to reuse training data, automatic methods for understanding rewards from images, and a system that helps robots reset tasks on their own. It also provides strong robot controllers to handle tricky tasks that involve physical contact.

SERL has been tested on real robots and has successfully learned tasks like assembling circuit boards, organizing objects, and routing cables—all in less than an hour of training. This challenges the common belief that RL is too slow for real-world use. Because SERL is open source, it makes advanced robotic learning more accessible to researchers and engineers.

By reviewing SERL’s features and results, this work highlights how it makes RL more practical and useful for robotics.

Faculty Mentor

Dr. Juan Rojas, juan.rojas@lipscomb.edu

Included in

Robotics Commons

Share

COinS
 

Making Robotic Reinforcement Learning More Efficient: Analyzing the SERL Framework

Teaching robots through reinforcement learning (RL) has made great progress, but real-world training is still difficult. Robots need lots of practice to learn, rewards can be hard to define, and resetting the environment after each attempt is often a challenge. The Sample-Efficient Robotic Reinforcement Learning (SERL) framework helps solve these issues by offering a ready-to-use, open-source software package that makes RL more practical for real-world robotics.

This project explores SERL and how it improves robotic RL by making learning faster and more efficient. SERL includes smarter ways to reuse training data, automatic methods for understanding rewards from images, and a system that helps robots reset tasks on their own. It also provides strong robot controllers to handle tricky tasks that involve physical contact.

SERL has been tested on real robots and has successfully learned tasks like assembling circuit boards, organizing objects, and routing cables—all in less than an hour of training. This challenges the common belief that RL is too slow for real-world use. Because SERL is open source, it makes advanced robotic learning more accessible to researchers and engineers.

By reviewing SERL’s features and results, this work highlights how it makes RL more practical and useful for robotics.

 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.