Back to Projects

Build from first principles

Build an RL Environment

Build a grid world an agent can learn in, watch the agent exploit a badly designed reward, fix the reward, and judge the results honestly.

What is an RL environment?Understand states, actions, rewards, and how an episode ends

1Move an Agent on a GridMove inside the world

2Decide the Reward and Whether the Episode EndedReward the agent and end episodes

3Compute the Learning TargetLearn from each step

4Score a Path with a Safe Progress BonusDesign a reward that cannot be farmed

5Summarize Runs Across SeedsJudge results honestly

From Pieces to a Working EnvironmentSee how your functions fit together before training an agent

Build and Break the EnvironmentTrain an agent, catch it gaming the reward, fix the reward, and compare across seeds