Generalization to Unseen Objects on MiniGrid Case 1 (test)
9.6Target Generalization ScoreFT (CONV.)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FT (CONV.)Evaluation Protocol=Fine-tuning (until convergence), Algorithm=DQN2026.04 | 9.6 | 0 | |
| ASPECTEvaluation Protocol=Zero-shot transfer, Semantic Mapper=MLLM (Grok-4.1-fast / Gemini 2.5 Flash)2026.04 | 9.4 | 0 | |
| SF-REC.Approach=Successor features with reconstruction2026.04 | 9 | 0 | |
| FT (LIM.)Evaluation Protocol=Fine-tuning (limited steps), Steps=50K, Algorithm=DQN2026.04 | 7.2 | 0.4 | |
| SF-SIMP.Approach=Successor feature-based2026.04 | 5.6 | 0 | |
| SOURCEEvaluation Protocol=Zero-shot transfer, Algorithm=DQN2026.04 | 0 | 0 |