Robots That Know When They Don't Know - Uncertainty Quantification in Machine Learning Systems

Watch a research seminar from Princeton University's Associate Professor Anirudha Majumdar exploring how robots can recognize their own uncertainty when encountering novel scenarios. Examine the challenges of using foundation models and machine learning systems in robotics, particularly focusing on perception and planning limitations. Learn about the critical need for robots to quantify uncertainty in their machine learning components, especially when dealing with unfamiliar objects or potentially unsafe LLM-generated plans. Discover approaches for developing more reliable autonomous systems that can accurately assess their own knowledge boundaries and respond appropriately when faced with uncertain situations.