Safe RL

A talk on reward design and safety in reinforcement learning, given at the “I Can’t Believe It’s Not Better” workshop at the Reinforcement Learning Conference.