Reinforcement Learning on 55 Atari games
1,160Mean Human-Normalized ScoreRAINBOW
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RAINBOWTraining frames=50M, Benchmark source=DQN-Zoo, Evaluation protocol=no-ops2023.10 | 1,160 | 154 | 37 | 52 | |
| PQRTraining frames=50M, Benchmark source=DQN-Zoo, Evaluation protocol=no-ops2023.10 | 1,121 | 124 | 33 | 53 | |
| RAINBOWTraining frames=50M, Benchmark source=Dopamine, Evaluation protocol=sticky2023.10 | 965 | 123 | 35 | 53 | |
| PQRTraining frames=50M, Benchmark source=Dopamine, Evaluation protocol=sticky2023.10 | 962 | 123 | 35 | 51 | |
| IQNTraining frames=50M, Benchmark source=Dopamine, Evaluation protocol=sticky, n-step updates=n=32023.10 | 940 | 124 | 32 | 51 | |
| IQNTraining frames=50M, Benchmark source=DQN-Zoo, Evaluation protocol=no-ops2023.10 | 902 | 131 | 31 | 50 | |
| QR-DQNTraining frames=50M, Benchmark source=Dopamine, Evaluation protocol=sticky2023.10 | 562 | 93 | 27 | 46 | |
| QR-DQNTraining frames=50M, Benchmark source=DQN-Zoo, Evaluation protocol=no-ops2023.10 | 559 | 118 | 29 | 47 | |
| DQNTraining frames=50M, Benchmark source=Dopamine, Evaluation protocol=sticky2023.10 | 401 | 51 | 15 | 0 | |
| DQNTraining frames=50M, Benchmark source=DQN-Zoo, Evaluation protocol=no-ops2023.10 | 314 | 55 | 18 | 0 |