Uncertainty Quantification for Learning-Enabled Robots

Explore a cutting-edge lecture on uncertainty quantification in learning-enabled robots presented by Anirudha Majumdar from Princeton University. Delve into the challenges of generalizing deep learning systems to novel scenarios in robotics, focusing on perception, planning, and natural language understanding. Discover innovative approaches to provide formal assurances for robots operating with rich sensory inputs and natural language instructions. Learn about the application of conformal prediction and generalization theory to complement foundation models, enabling robots to recognize their own limitations. Examine experimental validations of these methods, including LLM planners that request assistance when uncertain and vision-based navigation and manipulation systems with strong statistical guarantees. Gain insights from Majumdar's extensive research background and numerous accolades in the field of robotics and AI.