Reinforcement Learning on Q*bert Atari 2600 (test)
18,900Average Total RewardHuman
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Humanagent=expert human player2013.12 | 18,900 | — | |
| DQNpolicy=epsilon-greedy, epsilon=0.052013.12 | 1,952 | — | |
| Contingencypolicy=epsilon-greedy, epsilon=0.052013.12 | 960 | — | |
| Sarsapolicy=epsilon-greedy, epsilon=0.052013.12 | 614 | — | |
| Randompolicy=random2013.12 | 157 | — | |
| DQN Bestpolicy=epsilon-greedy, epsilon=0.05, evaluation=best episode2013.12 | — | 4,500 | |
| HNeat Bestpolicy=deterministic2013.12 | — | 1,800 | |
| HNeat Pixelpolicy=deterministic, input=8 color channel representation2013.12 | — | 1,325 |