Using Riemannian Optimization to learn Optimal Controllers
My paper focuses on a fundamental control problem called the linear-quadratic Gaussian using techniques from the emerging controls field called Direct Policy Optimization (DPO). DPO combines techniques from reinforcement learning and the theoretical safety guarantees from control theory to solve various control problems. I employed a novel technique called Riemannian optimization, which theoretically ensured convergence with an exceptionally fast rate. This new approach was an entire order of magnitude faster than the conventional policy gradient method.
I was awarded Best Student Paper at the 2024 IEEE Conference on Decision and Control amongst hundreds of student authors. The CDC is the largest and most prestigious international controls conference. This is a huge honor.