Beyond Rules and Rewards: From Pixels to Adaptive Policies Through Imitation, Reinforcement, and Multimodal Learning

Watch a comprehensive robotics colloquium lecture exploring advanced approaches in robotic learning, from data generation through imitation to multimodal systems integration. Discover how simulation-based training data enables robots to acquire generalizable skills through imitation, and learn about transformer-based policies that enhance robotic navigation and decision-making capabilities. Examine the integration of imitation learning with reinforcement learning to create adaptive behaviors for new tasks and environments. Explore techniques for policy generalization across multiple robot configurations and understand how vision-language models (VLMs) enable robots to interpret open-ended instructions for real-world applications. Presented by Kiana Ehsani, Senior Research Scientist at the Allen Institute for AI (PRIOR), this talk delves into cutting-edge research in Embodied AI, computer vision, and machine learning, demonstrating how these technologies are advancing robotic capabilities in dynamic and unstructured environments.