Q

Q-Learning Grid World

Tabular RL agent, trained live in 3D

Training

Hyperparameters

Place on grid

Click a tile to place. Click the same item again to remove it. Drag to orbit, scroll to zoom.

Environment

Display

Space start/pause · G greedy run · 1–6 tools

∑

Live Metrics

Idle · press Start Training

Epoch (episode)
0
Total steps
0
Exploration rate ε Exploring
1.000
Last epoch reward
-
Avg reward (50)
-
TD loss (MSE)
-
Goal rate (50)
-
Cumulative reward
0
Steps this epoch
0

Reward history

Epoch reward Moving avg (20) ε (0 to 1)

Heatmap · V(s) = maxa Q(s, a)

Low valueHigh value
Reach goal (ends epoch)+10 Coin (once per epoch)+3 Each step-0.1 Bump wall / edge-0.5 Hazard (ends epoch)-10

Event log