My research focuses on computational modeling and learning-based optimal control of complex dynamical systems, with applications in advanced manufacturing and robotics. A central goal is to develop models and controllers that retain relevant physical and mathematical structure while enabling real-time prediction and decision-making.
One research direction develops adjoint-based neural controllers that learn the sensitivity of long-term objectives to system states. These methods combine classical control theory, self-supervised learning, and online constrained optimization to compute control actions efficiently while enforcing actuator and safety constraints. Applications include nonholonomic robotic systems and, more broadly, nonlinear systems for which repeated online trajectory optimization is computationally demanding.
A second direction develops computationally efficient models of manufacturing processes. Projects include graph-based thermal modeling of directed energy deposition, diffusion-structured thermal surrogates for laser powder bed fusion, and neural surrogate modeling and latent-state inference for piezoelectric microdispensing. These projects combine numerical simulation and experimental measurements with scientific machine learning, system identification, and state estimation. Across these efforts, I investigate how physical structure can improve computational efficiency, predictive reliability, and the usefulness of learned models for optimization and feedback control.