Mechanical and Civil Engineering Seminar: PhD Thesis Defense

Abstract:

We cannot deploy the exciting capabilities of modern robotics without rigorous safety assurances. However, current approaches for achieving safety rely on conservative hardware designs or restrictive deployment regulations that severely limit these robots' capabilities.

To address this challenge, this talk proposes methods for achieving dynamic safety: formal safety guarantees that preserve system performance and remain valid under uncertainty. To this end, we will advance the theory and application of control barrier functions (CBFs), a leading framework for enforcing safety constraints on robotic systems. We begin by developing theoretical frameworks that extend CBF-based methods to settings with realistic uncertainties, and then incorporate machine learning techniques to improve performance while retaining robust safety assurances. Recognizing the complexity and randomness of the real world, we also propose an alternative, probabilistic understanding of safety and explore connections between CBFs and stochastic process theory that enable risk-sensitive control in uncertain settings. We then make these theoretical tools deployable and performant on compute-limited hardware systems by leveraging generative modeling tools and horizon-based control methods. Throughout this talk we will validate these methods with hardware demonstrations on a wide variety of robot platforms operating in complex environments under significant uncertainty.

Altogether, this work offers principled tools for deploying safe and capable robots, moving us closer to the goal of dynamic safety in the real world.

THURSDAY, MAY 29, 2025, 3:00 PM PT

Gates-Thomas 135, Hall Auditorium

or via Zoom: https://caltech.zoom.us/j/85154672791