Learning to Reject the Unknown: Disturbance Rejection, Learning, and Safety for Robotic Autonomy

Abstract

Autonomous robots are increasingly deployed in real-world environments where uncertainty is unavoidable. Wind gusts, modeling errors, unexpected contacts, and actuator faults can all degrade performance and potentially lead to failure. While modern learning-based methods have significantly advanced robotic capabilities, ensuring safety and reliability under unknown disturbances remains a central challenge for real-world autonomy.

This talk explores a control-theoretic approach augmented by data-driven methods for enabling robots to handle such uncertainties through learning-enabled disturbance rejection. I will present a framework that integrates disturbance estimation, machine learning, and safety-critical control to allow robotic systems to operate robustly in uncertain environments. In particular, I will discuss recent work on disturbance-rejection-guarded learning, where neural networks are combined with extended state observers to improve disturbance prediction while retaining robustness guarantees.

I will further show how disturbance-aware optimal control can be incorporated into safety-critical control frameworks to ensure constraint satisfaction and safe operation under uncertainty. These ideas will be demonstrated through applications in aerial robotics, robotic manipulation, and legged locomotion. Together, they point toward a new paradigm for robotic autonomy - systems that can not only learn from data, but also reliably and safely reject the unknown.

Bio: 

Dr. Qin Lin has been a tenure-track Assistant Professor in the Technology Division of the Cullen College of Engineering at the University of Houston since 2024. Prior to this, he was a tenure-track Assistant Professor in the Department of Computer Science at Cleveland State University from 2022 to 2024. He completed his postdoctoral training at the Robotics Institute of Carnegie Mellon University in 2021 and earned his Ph.D. in Computer Science from Delft University of Technology in 2019. Dr. Lin actively publishes papers in prestigious conferences and journals within the fields of robotics, control, and vehicle technology, including ACC, CDC, IFAC, IROS, ICRA, and various IEEE transactions. He serves as Associate Editors for esteemed journals such as IEEE Robotics and Automation Letters and IEEE Transactions on Vehicular Technology. His research is currently funded by grants from the NSF and the Department of Education.