Safe Planning for Embodied Agents on EMBGUARD (test)
95.1Safe PrecisionGemini-2.5-Pro
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini-2.5-Pro2026.05 | 95.1 | 42.8 | 33.4 | 54.9 | |
| Qwen-3-VL-32BModel Scale=32B2026.05 | 92.4 | 66 | 28.7 | 50 | |
| EMBGUARD-4BModel Scale=4B2026.05 | 92.1 | 61.5 | 23.4 | 40.4 | |
| GPT-5.12026.05 | 91.4 | 71 | 35.6 | 34.5 | |
| EMBGUARD-2BModel Scale=2B2026.05 | 90.2 | 40 | 6.6 | 40.8 |