Navigation on EmbodiedBench Vanilla
71.7Base Success RateSceneDiver (gemini-2.5-flash)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SceneDiver (gemini-2.5-flash)Model=gemini-2.5-flash, Configuration=Ours2026.06 | 71.7 | 67.7 | 61 | 61 | |
| gemini-2.5-flash (base)Model=gemini-2.5-flash, Configuration=Base2026.06 | 68.3 | 65 | 58.3 | 56.7 | |
| SceneDiver (Qwen2.5-VL-7B-Instruct-AWQ)Model=Qwen2.5-VL-7B-Instruct-AWQ, Configuration=Ours2026.06 | 40.7 | 29.7 | 38 | 32.3 | |
| Qwen2.5-VL-7B-Instruct-AWQ (base)Model=Qwen2.5-VL-7B-Instruct-AWQ, Configuration=Base2026.06 | 36.3 | 25 | 35 | 28.3 |