Plan Generation on 6 grid world domains Unseen Appearances
81.1Success Rate (Frozenlake)VLMFP
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VLMFPPlanning Approach=VLMFP, VLM Backbone=GPT-4o2025.10 | 81.1 | 82.8 | 25.1 | 53.4 | 61 | 21.3 | 54.1 | |
| CodePDDLPlanning Approach=CodePDDL, VLM Backbone=GPT-4o2025.10 | 77.1 | 74.9 | 0.4 | 19 | 22.1 | 0 | 32.3 | |
| CoTPlanning Approach=Chain-of-Thought, VLM Backbone=GPT-52025.10 | 28 | 16 | 13 | 3 | 12 | 0 | 12 | |
| DirectPlanning Approach=Direct, VLM Backbone=GPT-52025.10 | 24 | 15 | 9 | 3 | 9 | 0 | 10 | |
| CoTPlanning Approach=Chain-of-Thought, VLM Backbone=GPT-4o2025.10 | 10 | 2 | 0 | 0 | 0 | 0 | 2 | |
| DirectPlanning Approach=Direct, VLM Backbone=GPT-4o2025.10 | 8 | 2 | 0 | 0 | 0 | 0 | 1.7 |