The Reinforcement Learning Handbook: A Guide to Foundational Questions

Reinforcement Learning (RL) involves an agent learning to take actions within an environment to maximize cumulative rewards. The agent follows a policy, which is a strategy mapping states to actions, to interact with the environment and receive feedback. A fundamental challenge is the exploration-exploitation dilemma, where the agent must balance trying new actions to gather information against using its current best strategy. RL algorithms can be categorized as model-free, learning directly from experience, or model-based, where the agent first builds an internal simulation of the environment to plan its actions.