Tennis Pushing on VLA Backdoor Evaluation Tasks
98.2Success RateNone
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| NoneVLA Model=π02026.01 | 98.2 | — | |
| State BackdoorVLA Model=π02026.01 | 96.2 | 94.3 | |
| BadVLAVLA Model=π02026.01 | 93.4 | 65.9 | |
| NoneVLA Model=SmolVLA2026.01 | 92.3 | — | |
| NoneVLA Model=OpenVLA2026.01 | 92.1 | — | |
| State BackdoorVLA Model=OpenVLA2026.01 | 91.4 | 93.7 | |
| TrojanRobotVLA Model=π02026.01 | 90.7 | 58.9 | |
| BadVLAVLA Model=OpenVLA2026.01 | 89.5 | 59.1 | |
| NoneVLA Model=DP2026.01 | 89.4 | — | |
| State BackdoorVLA Model=SmolVLA2026.01 | 89.1 | 90.2 | |
| TrojanRobotVLA Model=OpenVLA2026.01 | 88.4 | 44.2 | |
| State BackdoorVLA Model=DP2026.01 | 88.3 | 91.5 | |
| TrojanRobotVLA Model=SmolVLA2026.01 | 87 | 35.8 | |
| BadVLAVLA Model=SmolVLA2026.01 | 86.3 | 54.4 | |
| TrojanRobotVLA Model=DP2026.01 | 85.3 | 50.4 | |
| BadVLAVLA Model=DP2026.01 | 84.2 | 61.8 | |
| NoneVLA Model=ACT2026.01 | 83.8 | — | |
| State BackdoorVLA Model=ACT2026.01 | 82.4 | 92.1 | |
| TrojanRobotVLA Model=ACT2026.01 | 81.2 | 38.7 | |
| BadVLAVLA Model=ACT2026.01 | 79.3 | 55.6 |