Loading...
Robot learning for lifelong adaptation in open-world environments
Siva, Sriram
Siva, Sriram
Citations
Altmetric:
Advisor
Editor
Date
Date Issued
2023
Date Submitted
Keywords
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
2024-10-18
Abstract
The use of robotic systems has become increasingly vital in modern society, owing to their superior precision, reliability, effectiveness, and flexibility in performing complex and demanding tasks, such as search and rescue operations, autonomous driving, planetary and subterranean exploration, human-robot collaboration, and homeland defense. With the integration of artificial intelligence and machine learning, the potential applications for robotic systems are expanding even further, enabling robots to perform tasks beyond their pre-programmed capabilities. As a result, robotic systems are now ubiquitous in our daily lives, providing greater convenience and ease of use, and offering new possibilities for innovation and growth.
Robot learning is critical for enabling intelligent robotic systems to operate successfully. By incrementally learning from their experiences or interactions with their environment, robots can improve their understanding of their surroundings and capabilities. However, in open-world environments, operations are conducted in highly unstructured, dynamic, and uncertain surroundings, posing significant challenges for robot learning. Accordingly, there is a need for robot learning that enables lifelong adaptation allowing robots to adjust their capabilities to continue to perform operations successfully and efficiently in open-world environments.
This dissertation focuses on addressing the challenges of robot lifelong operations in open-world environments by developing robot learning methods that enable adaptation on the fly. The research introduces regularized optimization-based and deep-learning methods that first make the robot robust and later enable it to adapt to long-term autonomy challenges with multiple sensor modalities. In addition, the dissertation proposes robot learning methods to adapt not only to changes in its surroundings but also within itself, such as changes in its capabilities. The proposed approaches have been successfully integrated into real-world robotic systems and demonstrate efficient solutions to real-world autonomy challenges through demonstrations in real-world operations.
Associated Publications
Rights
Copyright of the original work is retained by the author.