Data-driven methods for safety-critical control

Abstract: New automated systems with high levels of autonomy, such as self-driving cars and autonomous robots, are increasingly enabled by machine learning. These applications highlight the safety-critical nature of control systems technology, both due to their close proximity to the public and their level of autonomy. Safe control technology aims to provide guarantees that these systems will not do harm. Interest in safety filters, a modular approach to safe control, has increased in response to safety concerns associated with learning-based control often employed in robotics and autonomous driving. Such safety filters commonly rely on accurate mathematical models, contradicting the intended use to enhance data-driven learning solutions. This reliance on accurate models also limits the use of this technology in uncertain environments and in applications other than robotics. In this seminar I will highlight some of the challenges encountered when applying safe control to automated drug delivery. I will present recent results on data-driven safety filters that can extend the applicability of safe control <a href="http://technology.

Speaker(s):” target=”_blank” title=”technology.

Speaker(s):”>technology.

Speaker(s): Klaske van Heusden,

Room: SF 2104, Bldg: SF 2104, 172 St. George St., Toronto, Ontario, Canada, M5R 0A3