Social Navigation on Social Navigation Simulation 5 Human
100Success RateNaviWM
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NaviWMBackbone=GPT-52025.10 | 100 | 14.18 | 7.7 | 51 | 71 | |
| NaviWMBackbone=GPT-o42025.10 | 80 | 12.92 | 5.92 | 71 | 85 | |
| GPT-5Backbone=GPT-52025.10 | 60 | 11.97 | 4.65 | 62 | 89 | |
| NaviWMBackbone=GPT-4o2025.10 | 60 | 18.93 | 13.32 | 81 | 65 | |
| GPT-o4 (mini)Backbone=GPT-o4 (mini)2025.10 | 50 | 11.89 | 4.71 | 75 | 103 | |
| DeepSeek-V3Backbone=DeepSeek-V32025.10 | 40 | 14.68 | 7.08 | 22 | 37 | |
| Abl-WM-DeepSeekBackbone=DeepSeek2025.10 | 40 | 12.49 | 5.7 | 17 | 24 | |
| Abl-WM-GPTBackbone=GPT2025.10 | 40 | 11.35 | 8.4 | 31 | 49 | |
| GPT-4oBackbone=GPT-4o2025.10 | 30 | 6.95 | 3.08 | 41 | 45 | |
| Abl-CoT-GPTBackbone=GPT, Reasoning=Zero-Shot CoT2025.10 | 30 | 12.55 | 9.12 | 32 | 45 | |
| LLaMA-405BBackbone=LLaMA-405B2025.10 | 20 | 18.33 | 10.58 | 37 | 56 | |
| Abl-WM-LLaMABackbone=LLaMA2025.10 | 20 | 17.43 | 8.8 | 19 | 23 | |
| GPT-4Backbone=GPT-42025.10 | 10 | 15.21 | 25.3 | 50 | 87 | |
| LLaMA-8BBackbone=LLaMA-8B2025.10 | 0 | 27 | 21.78 | 7 | 13 | |
| GPT-3.5Backbone=GPT-3.52025.10 | 0 | 19.1 | 22.7 | 8 | 17 |