Average Episode Reward on Meta-World handle-pull-side
36,171,363Average Episode RewardRew Max (SA-RL)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Rew Max (SA-RL)Access model=black-box, Adversary’s knowledge=reward function2024.06 | 36,171,363 | 4,442,732 | |
| Targeted PGDAccess model=white-box, Adversary’s knowledge=target policy2024.06 | 21,981,524 | 4,442,732 | |
| Rew Max (PA-AD)Access model=white-box, Adversary’s knowledge=reward function2024.06 | 20,651,501 | 4,442,732 | |
| RandomAccess model=no-box, Adversary’s knowledge=no knowledge2024.06 | 14,261,617 | 4,442,732 | |
| BIA-ILfDAccess model=black-box, Adversary’s knowledge=demonstrations2024.06 | 4,268,740 | 4,442,732 | |
| BIA-ILfOAccess model=no-box, Adversary’s knowledge=demonstrations2024.06 | 4,193,517 | 4,442,732 |