Physical Red Teaming on RedVLA Conditional-Level Scenarios
98ASRπ0.5
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| π0.5Scenario=Environmental Harm2026.04 | 98 | 40 | |
| π0.5Scenario=Resource Damage (skull-marked)2026.04 | 96.7 | 0 | |
| OpenVLA-OFTScenario=Robot Damage2026.04 | 96.7 | 60 | |
| π0Scenario=Robot Damage2026.04 | 96.7 | 70 | |
| VLA-Adapter-ProScenario=Environmental Harm2026.04 | 94 | 40 | |
| OpenVLA-OFTScenario=Resource Damage (skull-marked)2026.04 | 93.3 | 0 | |
| VLA-Adapter-ProScenario=Resource Damage (skull-marked)2026.04 | 93.3 | 0 | |
| π0Scenario=Resource Damage (skull-marked)2026.04 | 93.3 | 0 | |
| OpenVLA-OFTScenario=Environmental Harm2026.04 | 92 | 36 | |
| π0.5Scenario=Dangerous Item Misuse2026.04 | 90 | 96.7 | |
| π0.5Scenario=Robot Damage2026.04 | 90 | 86.7 | |
| VLA-AdapterScenario=Environmental Harm2026.04 | 90 | 38 | |
| VLA-AdapterScenario=Robot Damage2026.04 | 86.7 | 70 | |
| VLA-Adapter-ProScenario=Robot Damage2026.04 | 83.3 | 80 | |
| VLA-AdapterScenario=Resource Damage (skull-marked)2026.04 | 80 | 0 | |
| π0Scenario=Dangerous Item Misuse2026.04 | 80 | 96.7 | |
| π0Scenario=Environmental Harm2026.04 | 78 | 40 | |
| OpenVLA-OFTScenario=Dangerous Item Misuse2026.04 | 73.3 | 86.7 | |
| VLA-Adapter-ProScenario=Dangerous Item Misuse2026.04 | 73.3 | 76.7 | |
| VLA-AdapterScenario=Dangerous Item Misuse2026.04 | 70 | 80 | |
| OpenVLAScenario=Environmental Harm2026.04 | 28 | 28 | |
| OpenVLAScenario=Dangerous Item Misuse2026.04 | 26.7 | 36.7 | |
| OpenVLAScenario=Robot Damage2026.04 | 26.7 | 26.7 | |
| OpenVLAScenario=Resource Damage (skull-marked)2026.04 | 23.3 | 0 |