Learning by Crashing Is Not an Option

Reinforcement Learning Agents Cannot Explore Safely in Physical Environments

digital-safe-rl-exploration-guarantees