Multi-task Reinforcement Learning on DMLab-30 Multi-task Standard
81.7Mean-Capped Human-Normalized ScoreLASER: Shared Experience Replay
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LASER: Shared Experience ReplayHyper-parameter sweep=true, Environment steps=10B2019.09 | 81.7 | 9,720 | |
| LASER: Experience ReplayHyper-parameter sweep=true, Environment steps=10B2019.09 | 79.6 | 9,540 | |
| PopArt-IMPALA+PixelControlHyper-parameter sweep=true, Environment steps=10B2019.09 | 77.6 | 8,550 | |
| PopArt-IMPALAHyper-parameter sweep=true, Environment steps=10B2019.09 | 73.5 | — |