Continuous Control on Ant (Robustness Gap)
0.16Robustness GapUP(mlp(wide))
Evaluation Results
| Method | Links | |
|---|---|---|
| UP(mlp(wide))Context=friction, State-action history length (k)=4, Data retention fraction (alpha)=Tuned2026.04 | 0.16 | |
| UP(mlp(narrow))Context=friction, State-action history length (k)=4, Data retention fraction (alpha)=Tuned2026.04 | 0.17 | |
| UP(lstm)Context=friction, State-action history length (k)=4, Data retention fraction (alpha)=Tuned2026.04 | 0.21 | |
| UP(gru)Context=friction, State-action history length (k)=4, Data retention fraction (alpha)=Tuned2026.04 | 0.21 | |
| UP(mlp(wide))Context=gravity, State-action history length (k)=4, Data retention fraction (alpha)=Tuned2026.04 | 0.45 | |
| UP(mlp(narrow))Context=gravity, State-action history length (k)=4, Data retention fraction (alpha)=Tuned2026.04 | 0.46 | |
| UP(gru)Context=gravity, State-action history length (k)=4, Data retention fraction (alpha)=Tuned2026.04 | 0.49 | |
| UP(lstm)Context=gravity, State-action history length (k)=4, Data retention fraction (alpha)=Tuned2026.04 | 0.51 |