Taewan Kim

Starting in October 2025, I will join SpaceX as a GNC Engineer.

I graduated from the University of Washington with a Ph.D. in Aeronautics & Astronautics, advised by Prof. Behçet Açıkmeşe. My doctoral work centers on control and optimization, where I enjoy both the theory behind control systems and the practical skills needed for high-performance optimization-based control.

Email  /  CV  /  Scholar  /  Linkedin  /  Github

Research

My work is in computational control. Specifically, I focus on formulating problems using robust control theory and computing trajectories, controlled invariant funnels, and feedback controllers using convex and nonconvex optimization methods. The list below contains some of my work. The full list is given in my Google Scholar.

Synthesis of trajectory, funnel, and feedback control

All about funnels.

Guaranteeing continuous-time invariance between discrete node points is challenging. This paper addresses the issue, and surprisingly, a matrix copositivity structure emerges.

Why not compute the trajectory itself as well? This paper does exactly that.

This reveals that funnel synthesis is essentially a trajectory optimization problem involving matrix variables. By applying numerical optimal control techniques, we can compute funnels that are less conservative compared to existing synthesis approaches.

Application of trajectory optimization for autonomous vehicles

Among the many nonconvex trajectory optimization methods, here is one tailored for the 6-DoF aircraft landing problem.

This work addresses trajectory planning for UAVs and their UGV carrier with operational constraints, such as UAV battery limitations and coordination requirements.

Policy Optimization Guided by Optimal Control

This work trains neural network policies for initializing trajectory optimization methods. The approach combines guided policy search with sequential convex programming for improved constraint satisfaction.

Training an end-to-end policy with guided policy search for target tracking. In hindsight, I don’t think end-to-end was the most suitable approach for this task, but it was an interesting exploration.

Website template from Jon Barron.