This will be an in-person event ONLY in Wu and Chen Auditorium.

ABSTRACT

Active sensing is often defined as “movement for the purpose of sensing.” Here, I take a different perspective—that active sensing in biological systems is not a distinct class of behaviors, but rather a set of movement phenomena that arise from a control objective. Biological sensors adapt to persistent stimuli, acting like high-pass filters that tend to block “DC.” Such “change-detecting” sensors can support efficient coding with a high dynamic range, and in engineering, bio-inspired event cameras are similar: they transmit information only when a pixel changes and, as such, are extremely fast and make efficient use of bandwidth for the right applications. However, such “AC” sensors pose technical challenges for control. Specifically, event-like biological sensors can cause a nonlinear system (1) to lose local linear observability and (2) to become impossible to stabilize about an equilibrium point (Biswas, Sontag, Cowan, Eur J Control, 2025). Active sensing behaviors must emerge for stable control, even in the somewhat paradoxical setting where the task-level goal is to remain stationary. Here, I will discuss my lab’s progress in analyzing how animals use active sensing behaviors to format sensory information, enhancing observability and control. I will also present our efforts to formalize controller synthesis with event-like sensors.