Reinforcement learning lets an agent improve its decisions by interacting with an environment: it observes a state, takes an action, receives a reward, and observes the next state. It is useful when the transition probabilities or rewards are not known in advance. section 6.5.3
Unlike dynamic programming with a known model, reinforcement learning estimates values from experience. It retains Bellman’s recursive logic: a decision’s value depends on its immediate reward and the value of what follows. Q-learning, for example, updates estimates for state-action choices as experience accumulates. section 6.5.3
Because decisions unfold in sequences, an agent can evaluate trajectories of states, actions, and outcomes. Their probabilities combine the conditional probabilities of successive transitions, supporting comparisons between strategies for maximizing long-term reward. section 2.4.2
Robot navigation
A robot in an unfamiliar environment learns without a model of where an action will lead or how rewarding the result will be; it uses observed outcomes to improve later decisions. section 6.5.3



























