Q-Learning

← Back to the article

Q-Learning learns by trying actions and remembering which ones lead to rewards.

What it does

Q-Learning learns by trying actions and remembering which ones lead to rewards.

Agent: The AI character • Reward: Points for good jobs

How it works

  • Try an action.
  • See the reward.
  • Update a table so better actions are easier to choose next time.

Uses and applications

  • A simple game AI can learn which moves win points.
  • A robot can learn which direction helps it reach a goal.
  • A learning app can choose practice questions that help most.

In short

It learns from rewards, like practice in a game.

Related methods

  • DQN: Replaces the Q-table with a neural network.
  • Multi-Armed Bandit: A simpler setting with rewards but no changing state.
  • MCTS (Tree Search): Chooses moves by searching ahead rather than learning a value table.

Back to Reinforcement learning, agents and planning · All topics

The interactive demonstration and the Python walkthrough for this topic need JavaScript.