Zih-Yun Chiu
University of California, San Diego
zchiu@ucsd.edu
Bio
Zih-Yun (Sarah) Chiu is a Ph.D. candidate in Electrical and Computer Engineering at the University of California San Diego (UCSD). She is advised by Professor Michael Yip. Her research focuses on high-precision robot autonomy for medical applications, addressing the sensing, planning, and adapting challenges in automation. She develops methods in probabilistic visual tracking, motion planning, and robot learning to enable autonomous agents to perform safely and efficiently in medical environments, including surgery and search-and-rescue scenarios. Her work has been published at robotics and ML venues, such as ICRA, IROS, and NeurIPS, and recognized with an ICRA Outstanding Healthcare and Medical Robotics Paper award. In addition, she interned at Amazon Robotics during her Ph.D. Before joining UCSD, she received her bachelor’s degree in Electrical Engineering from National Taiwan University.
Areas of Research
- Robotics
Safety and Efficiency in High-Precision Robot Autonomy for Medical Applications
Autonomous robotic interventions, in which robots independently perform medical operations such as surgeries and search-and-rescue, hold immense potential to reduce the burden on healthcare and rescue workers. Advancements in robotic techniques also open the possibility of increasing access to medical procedures in underserved and hazardous areas. To deliver effective treatment to humans, medical robots must be able to (1) balance precision and efficiency in real-time operations, (2) ensure anatomical and biomechanical safety during human-body manipulation, and (3) achieve generalizability across diverse environments, anatomies, and morphologies. However, the complexity of human bodies makes achieving accurate medical perception and quantifying human safety metrics challenging. Furthermore, robots’ struggle with rapid adaptability to unseen environments and continuous evolution makes medical automation less practical. My research focuses on sensing, planning, and adapting algorithms that enable robots to autonomously perform robust, safe, and efficient medical procedures. We first incorporate uncertainty, geometry constraints, and anatomy information into probabilistic visual tracking to achieve high-precision object localization in surgical environments. Our robot-sensing methods reach sub-millimeter accuracy in surgical environments and enable robots to succeed in automating real-world surgical tasks. Next, we propose mathematical models for environmental uncertainty and human biomechanical safety metrics and integrate them into constrained optimization frameworks to plan safe robotic trajectories. Our robot-planning frameworks achieved two groundbreaking advancements: increasing automation success by more than fourfold for surgical tasks and safely and precisely repositioning human bodies using a robotic manipulator. Finally, we developed reinforcement learning algorithms that flexibly leverage external guidance to enable efficient and life-long learning in medical applications and the general robot learning domain. Our robot-adapting algorithms achieve sample-efficient, generalizable, compositional, and incremental learning, which are the foundations for efficient robot learning.