Generalized Planning on Rovers
100ScaleExplicitly regularized Q-value policy (Ω^Exp.)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Explicitly regularized Q-value policy (Ω^Exp.)Architecture=R-GNN2026.03 | 100 | 66.5 | |
| Heuristically regularized Q-value policy (Ω^Heu.)Architecture=R-GNN2026.03 | 100 | 64.4 | |
| Heuristically regularized Q-value policy (Ω^Heu.)Architecture=OAE2026.03 | 61 | 29.8 | |
| Explicitly regularized Q-value policy (Ω^Exp.)Architecture=OAE2026.03 | 60 | 30.1 | |
| State-value policy (V)Architecture=R-GNN2026.03 | 53 | 29.8 | |
| Explicitly regularized Q-value policy (Ω^Exp.)Architecture=OE2026.03 | 48 | 24.6 | |
| Heuristically regularized Q-value policy (Ω^Heu.)Architecture=OE2026.03 | 46 | 23 | |
| Vanilla Q-value policy (Q)Architecture=R-GNN2026.03 | 36 | 17.1 | |
| Vanilla Q-value policy (Q)Architecture=OE2026.03 | 33 | 13.3 | |
| Vanilla Q-value policy (Q)Architecture=OAE2026.03 | 28 | 9.8 | |
| State-value policy (V)Architecture=OE2026.03 | 25 | 9.8 | |
| State-value policy (V)Architecture=OAE2026.03 | 22 | 8.3 |