Gridworld — Dynamic Programming

Reinforcement Learning — CentraleSupélec · Lecture 2: Dynamic Programming · Sutton & Barto Example 4.1

Preset Click on a cell to…

Inspect: show the Bellman backup of that state.

Discount γ = Reward per step = Reward for falling in a pit = Slip probability = Threshold θ

each move: reward terminal (T) pit: terminal with a penalty, reward wall changed / different selected state

Keys: ← → step, space play/pause.

Sutton & Barto, Reinforcement Learning: An Introduction (2nd ed.), §4.1–4.4 · Reinforcement Learning — CentraleSupélec