Juyeop Han

Juyeop Han is a PhD candidate in mechanical engineering whose research interests lie at the intersection of uncertainty quantification and machine learning. Specifically, he is interested in how quantifying uncertainty in learning models can aid in the development of safer, more efficient robotics systems. As a MathWorks Fellow, Juyeop will work on methods for uncertainty quantification (UQ) in neural networks and their integration into robotics applications. To that end, he has developed a novel visual-inertial odometry framework that incorporates a convolutional neural network for pose estimation, resulting in greater efficiency and accuracy compared to cases without UQ. His next steps include theoretically developing evidential learning methods for joint tasks of regression and classification, along with applying these methods to perception- aware control and planning. MathWorks has significantly facilitated his research, and he has contributed toolboxes for deep learning with UQ and for perception, planning, and control with uncertainty. Juyeop’s research has the potential to advance the field of robotics by leveraging evidential learning to enhance robot perception, control, and planning, thereby creating a safe and efficient software structure for robotics.