Reinforcement Learning on Atari 2600 GRAVITAR
3,351.4GRAVITAR ScoreAverage Human
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Average HumanPure-exploration regime=true2022.06 | 3,351.4 | — | |
| DQN MMCTraining frames=100 million, Sticky actions (s)=0.25, Frame skip=52018.07 | 1,078.3 | 251.1 | |
| ICMPure-exploration regime=true, Episodes=10, Seeds=32022.06 | 1,040.45 | — | |
| RNDPure-exploration regime=true, Episodes=10, Seeds=32022.06 | 954.84 | — | |
| BYOL-ExplorePure-exploration regime=true, Episodes=10, Seeds=32022.06 | 795.95 | — | |
| RNDTraining frames=100 million, Sticky actions (s)=0.25, Frame skip=52018.07 | 790 | 122.9 | |
| DQN MMC + SRTraining frames=100 million, Sticky actions (s)=0.25, Frame skip=52018.07 | 457.4 | 120.3 | |
| DQN MMC PIXELCNNTraining frames=100 million, Sticky actions (s)=0.25, Frame skip=52018.07 | 275.4 | — | |
| DQN MMC CTSTraining frames=100 million, Sticky actions (s)=0.25, Frame skip=52018.07 | 199.8 | — | |
| DQNTraining frames=100 million, Sticky actions (s)=0.25, Frame skip=52018.07 | 118.5 | 22 |