Xuan Wang, an assistant professor in Electrical and Computer Engineering at George Mason University’s College of Engineering and Computing, has received NSF funding for a project titled “CAREER: A Dynamical Systems Approach to Reliable and Efficient Knowledge Transfer in Robot Learning.” The grant supports research aimed at helping robots learn not just from data collected in the moment, but also by reusing prior experience when conditions change. That ability is essential for deploying robot learning in real-world settings where retraining can be expensive, slow, or risky.
Robot learning systems often stumble when they encounter a new task, environment, or hardware configuration. In many cases, they rely on large datasets and repeated trial-and-error to adapt, which limits their practicality when data collection is costly or unsafe. Wang’s work targets this bottleneck by focusing on how knowledge transfer can be made dependable under uncertainty.
The project draws on dynamical systems and control theory to study the mechanisms behind transfer. Rather than treating learning as a static mapping, the research examines how learned behaviors evolve over time and how internal dynamics can either preserve useful structure or cause transfer to fail. This perspective enables a sharper view of when adaptation will work reliably—and when it will break down unexpectedly.
A central goal is to develop methods that clarify the conditions that support efficient transfer. Wang will investigate how stability, robustness, and control-inspired constraints can be used to maintain performance while reducing the amount of new data required. By identifying failure modes early, the approach can reduce the need for extensive re-exploration when a robot moves into unfamiliar situations.
The team expects the results to inform new robot learning approaches that are both effective and data-efficient. Such techniques could accelerate the transition from laboratory demonstrations to field-ready autonomy. In practice, that means robots could adapt faster without demanding repeated, high-cost training cycles.
These advances align with national priorities for safer, affordable, and more reliable robotics. Potential applications include disaster response, healthcare support, environmental monitoring, manufacturing, logistics, and transportation—domains where reliability and efficiency are not optional.
For this research, Wang will receive $639,870 from the U.S. National Science Foundation. Funding begins in September 2026 and runs through late August 2031, supporting a multi-year effort to turn dynamical insights into actionable tools for robot intelligence.
Funded research like this reflects a broader push toward learning systems that generalize beyond their training conditions. By blending learning with dynamical systems thinking, the project aims to make knowledge transfer a predictable capability rather than an uncertain gamble.
Keywords
Robotics; robot learning; knowledge transfer; dynamical systems; control theory; data efficiency; reliability; NSF CAREER; stability; adaptation
