Continual Reinforcement Learning on CW10
92Success RateFT-MT-pre-trained
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FT-MT-pre-trainedSetup=Multi-task pre-trained2023.06 | 92 | 1,491.76 | — | |
| FT-MT-scratchSetup=Multi-task scratch2023.06 | 85 | 1,476.61 | — | |
| L2M-oracleVariant=Oracle2023.06 | 77 | 1,294.72 | 5 | |
| L2M2023.06 | 65 | 1,136.17 | 7 | |
| L2P-Pv2Variant=Pv22023.06 | 40 | 707.84 | 3 | |
| L2P-PreTVariant=PreT2023.06 | 34 | 606.33 | 1 | |
| L2P-PTVariant=PT2023.06 | 23 | 390.28 | 11 | |
| EWC2023.06 | 17 | 257.48 | 78 | |
| FT-last+headFine-tuned components=last layer and head2023.06 | 12 | 266.59 | 65 | |
| L22023.06 | 10 | 237.07 | 0 | |
| FTMode=Fine-tuning2023.06 | 9 | 234.57 | 75 | |
| FT-headFine-tuned components=head only2023.06 | 7 | 95.34 | 8 |