Explore cutting-edge developments in scaling safe learning to robotics in this 57-minute Winter 2021 Robotics Colloquium talk by Scott Niekum from the University of Texas, Austin. Delve into three key areas: a theory of safe imitation learning, scalable reward inference without models, and efficient policy evaluation. Discover how these algorithms blend safety and practicality to advance high-confidence robot learning with limited real-world data. Learn about the challenges of deploying learning robots in real-world scenarios and the importance of probabilistic guarantees for safety and performance. Gain insights from Niekum's expertise in imitation learning, reinforcement learning, and robotic manipulation as he discusses the progress towards making safe robot learning more practical and scalable.