2025-2026 Seminar Series
Upcoming Seminar
Speaker: Somil Bansal (Stanford University)
Title: Learning Safety in the Open World: From Reachability to Foundation Models
Abstract: As autonomous systems and robots are increasingly deployed in open-world settings, ensuring their safety has emerged as a fundamental challenge. Unlike controlled environments, open-world scenarios are characterized by pervasive uncertainty—in the environment, system dynamics, and, critically, in the semantics of what constitutes safe behavior. These challenges strain the assumptions underlying traditional safety frameworks rooted in control theory and formal verification.
In this talk, I will present a new perspective on extending classical safety paradigms to open-world autonomy. At the core is the idea of learning safety representations directly from rich, high-dimensional observations and human-interpretable specifications. Concretely, I will introduce observation-conditioned reachability, which enables safety reasoning directly from visual inputs, and language-conditioned reachability, which allows safety constraints to be specified and enforced through natural language. Building on these ideas, I will show how reachability-based safety representations can serve as real-time guardrails and fallback mechanisms for modern AI-enabled systems, including for state-of-the-art foundation and generative models.
Date: Friday, April 10th, 2026
Time: 9:00-10:00 AM PST
Link: https://dartmouth.zoom.us/j/94765993258?pwd=aErnkjFBqQaOTEgonxZkeNTbojHRbp.1
Line-Up and Video Links
| Date | Speaker | Title | YouTube |
|---|---|---|---|
| Dec. 11, 2025 | Heng Yang | Semidefinite Relaxations for Robot Perception and Control: From Theory to Practice and Back | (link) |
| Jan. 15, 2026 | Anirudha Majumdar | Video World Models That Know When They Don't Know | (link) |
| Feb. 19, 2026 | Thomas Lew | Uncertainty-Aware Control at the Limits | (link) |
| March 25, 2026 | Yan Gu | Prediction & Reaction in Motion: Humanoid Control for Non-Inertial Worlds and Athletic AI | (link) |
| April 10, 2026 | Somil Bansal | Learning Safety in the Open World: From Reachability to Foundation Models |