Offline Reinforcement Learning on HalfCheetah-v2
1.1Observation DeviationEDAC
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| EDACCorruption type=Adversarial, Corruption rate (c)=0.3, Corruption scale (epsilon)=1.02024.11 | 1.1 | 1.1 | -1.3 | -1.3 | |
| MSGCorruption type=Adversarial, Corruption rate (c)=0.3, Corruption scale (epsilon)=1.02024.11 | 1.1 | 37.3 | 47.7 | 47.7 | |
| UWMSGCorruption type=Adversarial, Corruption rate (c)=0.3, Corruption scale (epsilon)=1.02024.11 | 1.9 | 36.2 | -1.5 | -1.5 | |
| CQLCorruption type=Adversarial, Corruption rate (c)=0.3, Corruption scale (epsilon)=1.02024.11 | 5 | 32.6 | 43.8 | 42.6 | |
| IQLCorruption type=Adversarial, Corruption rate (c)=0.3, Corruption scale (epsilon)=1.02024.11 | 32.7 | -2.3 | -1.7 | -1.6 | |
| BCCorruption type=Adversarial, Corruption rate (c)=0.3, Corruption scale (epsilon)=1.02024.11 | 34.5 | 14 | 32.7 | 35.8 | |
| RIQLCorruption type=Adversarial, Corruption rate (c)=0.3, Corruption scale (epsilon)=1.02024.11 | 35.7 | 27.5 | 40.3 | 44.1 | |
| TRACERCorruption type=Adversarial, Corruption rate (c)=0.3, Corruption scale (epsilon)=1.02024.11 | 36.8 | 33.4 | 41.9 | 35.8 |