Robot Learning meets Formal Specifications: Designing Safer Embedded Software in the age of AI

Introduction There is considerable interest to use modules developed using deep learning in embedded software for complex cyber-physical systems (CPS) such as automobiles, aerial vehicles, and robots. Traditionally, such CPS applications have relied on human supervision for defining high-level plans, while using automated control techniques for lower level tasks; e.g., humans driving automobiles or pilots manually flying airplanes. Control engineering…

From the Editors: The First Year Report