CAREER: A Dynamical Systems Approach to Reliable and Efficient Knowledge Transfer in Robot Learning

This Faculty Early Career Development Program (CAREER) award supports research to improve the reliability, efficiency and interpretability of learning-based robot control. The project focuses on enabling robots to transfer and reuse prior knowledge when encountering new tasks, environments, or hardware platforms. Current robot learning methods often rely on large amounts of data and trial-and-error methods, limiting their use in conditions where data collection is costly, time-consuming, or unsafe. Drawing on concepts from dynamical systems and control, this project will study how knowledge transfer occurs, when it breaks down, and how it can be made more efficient. The resulting advances will support new robot learning approaches that are both effective and data-efficient. These outcomes align with national priorities by enabling safer, more affordable, and more reliable robotic technologies in areas such as disaster response, healthcare support, environmental monitoring, manufacturing, logistics, and transportation. Closely integrated with the research, the project will advance education and workforce development by creating hands-on learning opportunities that connect foundational concepts with real-world robotic systems. Outreach efforts will expand participation in control, robotics, and artificial intelligence through activities such as a RoboArt program that encourages creative robot design for K–8 students; research experiences for high school, undergraduate, and graduate students; and new course materials that prepare students for leadership roles in emerging technology sectors. The project models knowledge transfer in robot learning as a principled dynamical system, and develops formal methods to analyze and regulate its behavior. The project consists of three main components. First, it introduces structural modularization to describe the knowledge transfer process of a single-robot as a dynamical system, enabling formal analysis of stability and convergence. Second, it extends the framework to multi-robot systems by modeling robot-to-robot knowledge sharing as a distributed dynamical system over time-varying interaction graphs, allowing analysis of collective learning behavior. Third, it develops new methods to regulate the transfer process through targeted knowledge augmentation and robust fusion strategies that reduce the impact of corrupted or misleading information. These methods draw on tools from control theory, machine learning, optimization, and networked systems, and will be validated through simulation and experiments on heterogeneous wheeled and legged robots performing tasks such as environmental search and cooperative transportation. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.