Reward function

Definition

In reinforcement learning, a numerical signal that tells the agent, at each step, whether what it has just done brings it closer to the goal. The agent seeks to maximise the sum of the rewards received over time, not only the immediate reward. The reward tells the agent what to achieve, not how to achieve it.

Checked on 28 September 2026 against the references below.

Related terms

• Reinforcement learning
• Policy

References

• Sutton and Barto, Reinforcement Learning: An Introduction, § 1.3 Elements of Reinforcement Learning
• Sutton and Barto, Reinforcement Learning: An Introduction, § 3.2 Goals and Rewards