Ya-Chien Chang

University of California San Diego, University of San Diego

Papers

Total Citations

H-Index

About

Ya-Chien Chang is a leading researcher at the intersection of reinforcement learning, control theory, and robotics, with a core focus on bridging the gap between data-driven methods and formal safety guarantees. Her most significant contribution is pioneering the use of neural Lyapunov functions to stabilize neural network-based controllers, directly addressing the critical lack of stability guarantees that has long hindered the practical deployment of learning-based methods in robotics. Her seminal 2021 work, "Stabilizing Neural Control Using Self-Learned Almost Lyapunov Critics," which has garnered 34 citations, introduced a novel model-free reinforcement learning framework that simultaneously learns a control policy and a neural Lyapunov critic, ensuring provable stability without requiring a system model. Building on this, her 2023 work on "Learning Stabilization Control from Observations" extends these stability principles to the challenging Learning from Observations (LfO) paradigm, enabling robots to learn stabilizing controllers directly from expert state trajectories without hand-crafted reward functions. This work is particularly impactful for real-world applications where reward engineering is difficult. Chang’s research is fundamentally reshaping how we think about safe, verifiable learning for autonomous systems.