High-performance cyber-physical systems rely on many sensors and hardware components for successful operation. Control strategies for these devices require an understanding of how unpredictability in these components might impair performance. In this talk, we aim to quantify the informational bottlenecks imposed by uncertain system models. We can use this to quantify the value of side-information regarding the uncertainty in the system (in bits), in order to answer questions such as: "what is the value of adding an extra sensor to the system?" We will also show that systems with uncertain actuation (e.g., when motors on a drone cannot precisely execute control actions) exhibit surprisingly different behavior than systems with uncertain sensing (e.g., miscalibrated cameras).

The talk will include joint work with Jian Ding, Yuval Peres, Govind Ramnarayan, Anant Sahai, and Alex Zhai.