Skip to content
Back to course

16.2 Rewards

A reward is a single number the environment gives the agent after each action, signaling how good that action was. The agent's whole goal is to maximize the cumulative reward over time, not just the immediate reward. This long-term total is often called the return.

Imagine training a delivery rider in Addis: giving +10 for every package delivered on time and -1 for each minute late teaches the rider to balance speed with reliability, instead of only chasing one quick win.

Scenario

You want a cleaning robot to fully clean a room. Which reward design is safest?

Check your understanding

1/4 · 40 XP

What is the agent's true objective in RL?