Tyler Han

I am currently a PhD student in the Paul G. Allen School of Computer Science & Engineering at the University of Washington. I am part of the Robot Learning Lab where I am advised by Byron Boots. I am also an NSF Graduate Research Fellow.

Prior to UW, I completed my B.S. in Aerospace Engineering and B.S. in Computer Science at the University of Maryland, College Park. During my undergrad, I worked with Glen Henshaw and Patrick Wensing while at the Naval Research Laboratory in Washington, D.C.

Email  /  GitHub  /  LinkedIn

Research

I am generally interested in techniques that combine machine learning and engineering. Within robotics, I am particularly interested in ill-defined engineering and autonomy challenges. Often, these problems require some "human intuition" to help constrain the solution. Some techniques I have been investigating involve learning from demonstration (LfD) and Inverse Reinforcement Learning (IRL).

Miscellaneous Projects

Other projects that are not necessarily research related or not published

Off-Road Self-Driving

Since beginning my PhD, I have been primarily focused on work related to the DARPA RACER program as part of the UW team. The robot itself is a modified Polaris RZR with onboard sensing and compute. My work on the team has consisted of a range of jobs: conducting field tests (experimental runs with new perception, planning, and/or control methods), research and development of optimal control algorithms, developing better robot software infrastructure, and more. I am now focused on improving the autonomous behavior through imitation and reinforcement learning.

Learning Motor Primitives

In an undergraduate project, I tackled part of the challenge of teaching robots to perform motor skills from a small number of demonstrations. We proposed a novel approach by joining the theories of Koopman Operators and Dynamic Movement Primitives to Learning from Demonstration. Our approach, named Autoencoder Dynamic Mode Decomposition (aDMD), projects nonlinear dynamical systems into linear latent spaces such that a solution reproduces the desired complex motion. Use of an autoencoder in our approach enables generalizability and scalabil- ity, while the constraint to a linear system attains interpretability. We show results on the LASA Handwriting dataset but with training on only a small fractions of the letters.

Forked from Leonid Keselman's Website