| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Fine-grained action recognition | Gym99-skeleton severe temporal corruption (Sev.) | Top-1 Acc89.1 | 9 | |
| Fine-grained action recognition | Gym99-skeleton (moderate temporal corruption (Mod.)) | Top-1 Acc90.6 | 9 | |
| Fine-grained action recognition | Gym99-skeleton minor temporal corruption (Min.) | Top-1 Acc92.1 | 9 | |
| Fine-grained action recognition | Gym288-skeleton (severe temporal corruption) | Top-1 Acc78.1 | 9 | |
| Fine-grained action recognition | Gym288-skeleton moderate temporal corruption (Mod.) | Top-1 Acc79.7 | 9 | |
| Fine-grained action recognition | Gym288-skeleton minor temporal corruption (Min.) | Top-1 Acc81.5 | 9 | |
| Continuous Control | Gym MuJoCo | Normalized Reward (TD3)1.54 | 8 | |
| Video Recognition | GYM | Accuracy90.8 | 8 | |
| Offline Reinforcement Learning | Gym Locomotion (Overall) | Normalized Score81.8 | 8 | |
| Action Classification | Gym288 | Accuracy (Per-Class)61.9 | 8 | |
| Action Classification | Gym99 | Per-Class Accuracy90.6 | 8 | |
| Skeleton Restoration | Gym99-Skeleton (minor) | MPJPE0.106 | 6 | |
| Aggregate Efficiency | GYM | Runtime Ratio (vs DHMBPO)1 | 3 | |
| Walker2d | GYM | Runtime (hours)4 | 3 | |
| Humanoid | GYM | Runtime (hours)5.2 | 3 | |
| Hopper | GYM | Runtime (hours)3.7 | 3 | |
| HalfCheetah | GYM | Runtime (hours)3.3 | 3 | |
| Ant | GYM | Runtime (hours)3.6 | 3 | |
| Lifelong Reinforcement Learning | Gym Control | CartPole Score39.6 | 3 |