Social Navigation on Social Navigation Simulation 10 Human
80Success Rate (SR)NaviWM
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NaviWMBackbone=GPT-52025.10 | 80 | 14.99 | 8.23 | 72 | 185 | |
| NaviWMBackbone=GPT-4o2025.10 | 60 | 20.15 | 16.6 | 87 | 173 | |
| NaviWMBackbone=GPT-o42025.10 | 60 | 12.19 | 7.45 | 83 | 171 | |
| GPT-o4 (mini)Backbone=GPT-o4 (mini)2025.10 | 50 | 11.63 | 6.63 | 99 | 149 | |
| GPT-5Backbone=GPT-52025.10 | 40 | 11.45 | 6.5 | 87 | 170 | |
| Abl-WM-DeepSeekBackbone=DeepSeek2025.10 | 40 | 10.35 | 4.58 | 15 | 33 | |
| DeepSeek-V3Backbone=DeepSeek-V32025.10 | 30 | 9.6 | 3.68 | 28 | 47 | |
| GPT-4oBackbone=GPT-4o2025.10 | 30 | 17.74 | 7 | 70 | 115 | |
| Abl-WM-GPTBackbone=GPT2025.10 | 30 | 10.28 | 3.58 | 21 | 37 | |
| Abl-WM-LLaMABackbone=LLaMA2025.10 | 20 | 12.93 | 5.55 | 20 | 45 | |
| Abl-CoT-GPTBackbone=GPT, Reasoning=Zero-Shot CoT2025.10 | 20 | 8.79 | 3.6 | 30 | 59 | |
| LLaMA-8BBackbone=LLaMA-8B2025.10 | 0 | 26.84 | 20.5 | 10 | 23 | |
| LLaMA-405BBackbone=LLaMA-405B2025.10 | 0 | 7.19 | 3.25 | 26 | 52 | |
| GPT-3.5Backbone=GPT-3.52025.10 | 0 | 0 | 43.25 | 34 | 82 | |
| GPT-4Backbone=GPT-42025.10 | 0 | 15.29 | 27.6 | 69 | 139 |