Preparation tasks on Habitat-Matterport 3D (HM3D)
19.5Success RateReflective Test-Time Planning
Evaluation results
| Method | Links | |
|---|---|---|
| Reflective Test-Time PlanningReflection-in-action=true, Reflection-on-action=true, Action Model Loss=true, Internal Reflection Model Loss=true, Setup=Zero-shot transfer2026.02 | 19.5 | |
| Reflective Test-Time PlanningAction Model Loss=false, Setup=Zero-shot transfer2026.02 | 14.6 | |
| Reflective Test-Time PlanningInternal Reflection Model Loss=false, Setup=Zero-shot transfer2026.02 | 9.76 | |
| 3DLLM-MemSetup=Zero-shot transfer2026.02 | 7.32 | |
| Reflective Test-Time PlanningReflection-in-action=true, Reflection-on-action=false, Setup=Zero-shot transfer2026.02 | 7.32 | |
| Self-RefineSetup=Zero-shot transfer2026.02 | 4.88 | |
| ReflexionSetup=Zero-shot transfer2026.02 | 2.44 | |
| DreamerV3Setup=Zero-shot transfer2026.02 | 2.44 | |
| Reflective Test-Time PlanningReflection-in-action=false, Reflection-on-action=true, Setup=Zero-shot transfer2026.02 | 2.44 | |
| ReflectVLMSetup=Zero-shot transfer2026.02 | 0 | |
| PPOSetup=Zero-shot transfer2026.02 | 0 | |
| Reflective Test-Time PlanningReflection-in-action=false, Reflection-on-action=false, Setup=Zero-shot transfer2026.02 | 0 |