Best Attack Reward on Meta-World v2 (handle-press-side, test)
4,726Best Attack RewardPPO
Evaluation Results
| Method | Links | |
|---|---|---|
| PPODefense mechanism=No defense2024.06 | 4,726 | |
| ATLA-PPODefense mechanism=AdvTraining2024.06 | 4,302 | |
| PA-ATLA-PPODefense mechanism=AdvTraining2024.06 | 3,318 | |
| WocaR-PPODefense mechanism=Partial smoothing2024.06 | 3,042 | |
| RAD-PPODefense mechanism=Regret2024.06 | 2,375 | |
| TDRT-PPODefense mechanism=Smoothing, Time-discounting=true2024.06 | 1,928 | |
| SA-PPODefense mechanism=Smoothing, Time-discounting=false2024.06 | 1,888 |