Embodied Planning on ALFWorld Text Aligned (134 unseen tasks)
98.5Success RateEvoPlan
Evaluation results
| Method | Links | ||
|---|---|---|---|
| EvoPlanBackbone=Qwen3-32B2026.07 | 98.5 | 853,022 | |
| NL-PDDL2026.07 | 94 | 443,407 | |
| ReflexionPlanner Type=Reflective LLM-based, Iterations=102026.07 | 91 | — | |
| ReflexionPlanner Type=Reflective LLM-based, Iterations=32026.07 | 83 | — | |
| ReActPlanner Type=Reflective LLM-based, In-context Examples=with examples2026.07 | 81 | 5,507,266 | |
| DEPSPlanner Type=Reflective LLM-based2026.07 | 76 | — | |
| ReActPlanner Type=Reflective LLM-based, In-context Examples=with model2026.07 | 34 | 4,242,560 | |
| BUTLERFine-tuning=100K expert traj.2026.07 | 26 | — | |
| GPT-4oPlanner Type=Direct LLM/VLM-based, Backbone=GPT-4o2026.07 | 21 | 1,358,934 | |
| LLaMA-3.1Planner Type=Direct LLM/VLM-based, Backbone=LLaMA-3.12026.07 | 19 | 1,193,640 | |
| Gemini-2.0 FlashPlanner Type=Direct LLM/VLM-based, Backbone=Gemini-2.0 Flash2026.07 | 16 | 1,176,406 |