Button Pressing on VLA Backdoor Evaluation Tasks
0.984Success RateNone
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| NoneVLA Model=π02026.01 | 0.984 | — | |
| State BackdoorVLA Model=π02026.01 | 0.964 | 0.953 | |
| BadVLAVLA Model=π02026.01 | 0.931 | 0.708 | |
| State BackdoorVLA Model=OpenVLA2026.01 | 0.921 | 0.946 | |
| NoneVLA Model=OpenVLA2026.01 | 0.92 | — | |
| TrojanRobotVLA Model=π02026.01 | 0.913 | 0.602 | |
| NoneVLA Model=SmolVLA2026.01 | 0.911 | — | |
| BadVLAVLA Model=OpenVLA2026.01 | 0.906 | 0.652 | |
| State BackdoorVLA Model=SmolVLA2026.01 | 0.893 | 0.904 | |
| TrojanRobotVLA Model=OpenVLA2026.01 | 0.891 | 0.498 | |
| NoneVLA Model=DP2026.01 | 0.883 | — | |
| State BackdoorVLA Model=DP2026.01 | 0.875 | 0.925 | |
| TrojanRobotVLA Model=SmolVLA2026.01 | 0.862 | 0.445 | |
| BadVLAVLA Model=SmolVLA2026.01 | 0.859 | 0.631 | |
| NoneVLA Model=ACT2026.01 | 0.846 | — | |
| TrojanRobotVLA Model=DP2026.01 | 0.844 | 0.583 | |
| State BackdoorVLA Model=ACT2026.01 | 0.837 | 0.942 | |
| BadVLAVLA Model=DP2026.01 | 0.834 | 0.667 | |
| TrojanRobotVLA Model=ACT2026.01 | 0.821 | 0.456 | |
| BadVLAVLA Model=ACT2026.01 | 0.808 | 0.612 |