Jasmine Terrones

Jasmine Terrones is a graduate student in mechanical engineering investigating how Koopman operator theory enables the use of classical control methods for systems with contact dynamics, such as legged robots. Legged robots—whose governing equations switch depending on which legs touch ground—resist traditional control approaches. Jasmine develops globally valid linear time-invariant models of these switched nonlinear dynamics, enabling rapid convex optimization through contact modes. She constructs all simulation data in MATLAB, deriving equations of motion symbolically with the Symbolic Toolbox. Once a model is trained, Koopman model predictive control can actuate a legged robot without a predefined contact sequence and find novel control strategies whose benefits are only realized at future contact modes. As a MathWorks Fellow, Jasmine will extend Koopman modeling with uncertainty quantification to improve controller performance using MATLAB’s Control System Toolbox. She also plans to develop novel algorithms for handling stiff dynamics. These methods could enable safe and fast control in complex environments where traditional control techniques fundamentally break down.