Decision Making on MINIGRID
60Success RateReAct with ChatDB memory
Evaluation Results
| Method | Links | |
|---|---|---|
| ReAct with ChatDB memoryReasoning=ReAct, Memory=ChatDB, Reflection=None, Backbone=Qwen3.5-27B2026.06 | 60 | |
| MAD with Base memoryReasoning=MAD, Memory=Base, Reflection=None, Backbone=GPT-5 mini2026.06 | 60 | |
| MAD with MemoryBank and ReflexionReasoning=MAD, Memory=MemoryBank, Reflection=Reflexion, Backbone=Qwen3.5-27B2026.06 | 60 | |
| ReAct with Base memoryReasoning=ReAct, Memory=Base, Reflection=None, Backbone=Qwen3.5-27B2026.06 | 50 | |
| ReAct with MemoryBank and ReflexionReasoning=ReAct, Memory=MemoryBank, Reflection=Reflexion, Backbone=Qwen3.5-2B2026.06 | 50 | |
| None reasoning with Base memoryReasoning=None, Memory=Base, Reflection=None, Backbone=Qwen3.5-27B2026.06 | 40 | |
| CoT with SimpleMem and ReflexionReasoning=CoT, Memory=SimpleMem, Reflection=Reflexion, Backbone=Qwen3.5-0.8B2026.06 | 40 | |
| CoT with Base memory and Self-RefineReasoning=CoT, Memory=Base, Reflection=Self-Refine, Backbone=Qwen3.5-0.8B2026.06 | 30 | |
| Plan&Solve with Base memory and ReflexionReasoning=Plan&Solve, Memory=Base, Reflection=Reflexion, Backbone=Qwen3.5-9B2026.06 | 20 |