Reinforcement Learning on Atari 2600 (Comprehensive Game Set)
7,127.7Alien ScoreHuman
Evaluation results
| Method | Links | ||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| HumanEvaluation Protocol=Maximal evaluation scores2016.02 | 7,127.7 | 47,388.7 | 753.1 | 16,926.5 | 160.7 | 30.5 | 860.5 | 29.6 | 4,334.7 | 4,753.3 | 14.6 | 13,455 | 42,054.7 | 1,668.7 | — | 1,719.5 | 742 | 8,503.3 | 29,028.1 | 37,187.5 | 12.1 | 12,017 | 7,387.8 | 35,829.4 | 1,971 | -16.4 | -38.7 | 2,412.5 | 3,351.4 | 30,826.4 | 0.9 | 302.8 | 3,035 | 2,665.5 | 22,736.3 | 6,951.6 | 8,049 | 69,571.3 | 17,118 | 7,845 | 11.9 | 10,250 | -8.3 | 5,229.2 | 167.6 | 11,693.2 | 1,187.5 | 17,667.9 | 4,756.5 | 9,173.3 | |
| DDQNEvaluation Protocol=Maximal evaluation scores2016.02 | 4,007.7 | 1,981.4 | 1,109 | 16,650.7 | 77.9 | 437 | 1,496.7 | 33.4 | 2,766.8 | 0 | 20.9 | 15,209.7 | 19,183.9 | 4,715.8 | — | 2,138.3 | 6,997.9 | 17,366.4 | 767,850 | 34,620.7 | 90.2 | 4,855.4 | 5,019 | 137,244.4 | 98,450 | -1.8 | 19.8 | 13,815.9 | 708.6 | 20,974.2 | -1.7 | 1,120.2 | 14,717.6 | 9,690.9 | 36,365.7 | 3,424.6 | 11,744.4 | 158.4 | 14,555.1 | 49,518.4 | 70.6 | 66,091.2 | 11.8 | 10,075.8 | 268 | 19,743.5 | 239.7 | 685,911 | 7,655.7 | 12,947.6 | |
| NatureEvaluation Protocol=Maximal evaluation scores2016.02 | 3,069 | 1,629 | 429.7 | 6,846 | 42.4 | 401.2 | 301.8 | 30.3 | 328.3 | 0 | 18.9 | 10,596 | 5,286 | 1,976 | — | 739.5 | 3,359 | 6,012 | 85,641 | 26,300 | 71.8 | 8,309 | 6,687 | 114,103 | 9,711 | -18.1 | -0.8 | 8,520 | 306.7 | 19,950 | -1.6 | 576.7 | 6,740 | 3,805 | 23,270 | 2,311 | 7,257 | 1,788 | 8,316 | 18,257 | 51.6 | 57,997 | -2.5 | 5,947 | 186.7 | 8,456 | 380 | 42,684 | 3,393 | 4,977 | |
| Bootstrapped DQNEvaluation Protocol=Maximal evaluation scores2016.02 | 2,436.6 | 1,032 | 1,208 | 23,429.8 | 60.2 | 855 | 1,591 | 33.9 | 2,181.4 | 100 | 20.9 | 15,092.7 | 9,083.1 | 2,893 | — | 1,272.5 | 8,047.1 | 19,713.2 | 994,500 | 38,666.7 | 93.2 | 4,553.5 | 4,100 | 137,925.9 | 82,610 | 3 | 26 | 17,438.4 | 286.1 | 21,021.3 | -1.3 | 1,663.5 | 14,862.5 | 8,627.9 | 36,733.3 | 2,983.3 | 11,501.1 | 1,812.5 | 12,845 | 51,500 | 66.6 | 55,725 | 0 | 9,079.4 | 214.8 | 26,231 | 212.5 | 811,610 | 6,804.7 | 11,491.7 | |
| RandomEvaluation Protocol=Maximal evaluation scores2016.02 | 227.8 | 719.1 | 14.2 | 363.9 | 23.1 | 1.7 | 0 | 0 | 65.2 | 0 | -20.7 | 163.9 | 68.4 | 148 | — | 5.8 | 222.4 | 210 | 12,850 | 2,360 | 0.1 | 2,090.9 | 811 | 10,780.5 | 152.1 | -18.6 | -91.7 | 257.6 | 173 | 1,027 | -11.2 | 29 | 52 | 1,598 | 258.5 | 307.3 | 2,292.3 | 24.9 | 1,338.5 | 11.5 | 2.2 | 664 | -23.8 | 3,568 | 11.4 | 533.4 | 0 | 0 | 563.5 | 32.5 | |
| Boot-DQN+head rescaling=false2016.02 | 33 | 55 | 77 | 79 | 54 | 52 | 172 | 81 | 98 | 0 | 60 | 24 | 44 | 38 | 62 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DQN2016.02 | 23 | 29 | 6 | 24 | 24 | 6 | 168 | 58 | 99 | 0 | -13 | 13 | 18 | 25 | 35 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Boot-DQNgradient normalization=true2016.02 | 23 | 29 | 9 | 46 | 56 | 16 | 185 | 68 | 112 | 0 | 2 | 16 | 23 | 30 | 41 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Thompson Sampling2015.07 | 20.4 | 22.3 | 13.03 | 8.97 | 112.2 | 25.4 | 46.6 | 22.8 | 74.6 | 0 | 61.2 | 12.7 | 17.4 | 14.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Dynamic AE2016.02 | 20 | 41 | 15 | 9 | 149 | 20 | 49 | 21 | 97 | 0 | 56 | 10 | 17 | 18 | 37 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Exploration Static AE2015.07 | 19.8 | 41.5 | 14.59 | 9.19 | 149.3 | 20.2 | 49.5 | 21.3 | 97.1 | 0 | 56 | 10.4 | 17.2 | 18.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Boltzman Exploration2015.07 | 18.7 | 45.6 | 8.9 | 8.17 | 133.8 | 29.4 | 53.8 | 31.3 | 92.8 | 0 | 61.2 | 13 | 19.4 | 18.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Exploration Dynamic AE2015.07 | 17.1 | 25.4 | 8.9 | 11.12 | 183.6 | 19.2 | 58.9 | 29.5 | 62.2 | 0 | 42.4 | 12.1 | 26.5 | 21.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DQN2015.07 | 15.3 | 25.9 | 7.15 | 11.22 | 96.4 | 19.1 | 51.8 | 20.6 | 57.3 | 0 | 52 | 15.5 | 16 | 20.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DQN*2016.02 | 15 | 26 | 7 | 11 | 96 | 19 | 52 | 21 | 57 | 0 | 52 | 15 | 16 | 20 | 29 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |