Reinforcement Learning Simulator

Q-learning · rewards · exploration vs exploitation
STATUS READY
EPISODE 0
MAP MAZE

Q-learning arena

Two agents learn independently from rewards. Watch exploration become a useful policy over repeated episodes.

CURRENT MAP · MAZE
BLUE AVG 0.0
ORANGE AVG 0.0
Click/drag to edit while paused
Recent episode reward · last 50 episodes
BlueOrange
StartWallLava −10Goal +10
Policy arrows show the highest learned Q-value action(s) in each state.