Robotic Manipulation (Object) on LIBERO-Object 1.0 (test)
98.7Success Rate (SR)AutoHorizon
Evaluation results
| Method | Links | ||
|---|---|---|---|
| AutoHorizonBackbone=GR00T N1.52026.02 | 98.7 | — | |
| Static OracleExecution Horizon (e)=8, Backbone=GR00T N1.52026.02 | 97.3 | — | |
| Static Oracle+Backbone=GR00T N1.52026.02 | 97.3 | — | |
| Static OracleExecution Horizon (e)=12, Backbone=GR00T N1.52026.02 | 96 | — | |
| RandomBackbone=GR00T N1.52026.02 | 96 | — | |
| Static OracleExecution Horizon (e)=2, Backbone=GR00T N1.52026.02 | 95.3 | — | |
| Static OracleExecution Horizon (e)=4, Backbone=GR00T N1.52026.02 | 95.3 | — | |
| Static OracleExecution Horizon (e)=1, Backbone=GR00T N1.52026.02 | 94.7 | — | |
| Static OracleExecution Horizon (e)=16, Backbone=GR00T N1.52026.02 | 94.7 | — | |
| TrojanRobotVLA Model=DP2026.01 | 89.4 | 83.2 | |
| State BackdoorVLA Model=DP2026.01 | 89.3 | 88.2 | |
| NoneVLA Model=ACT2026.01 | 88.4 | — | |
| BadVLAVLA Model=π02026.01 | 88.2 | 88.7 | |
| TrojanRobotVLA Model=SmolVLA2026.01 | 88 | 80.6 | |
| BadVLAVLA Model=DP2026.01 | 87.5 | 85.4 | |
| NoneVLA Model=π02026.01 | 87 | — | |
| State BackdoorVLA Model=ACT2026.01 | 86.2 | 85.5 | |
| TrojanRobotVLA Model=ACT2026.01 | 86.1 | 84.7 | |
| State BackdoorVLA Model=SmolVLA2026.01 | 85.5 | 88 | |
| BadVLAVLA Model=SmolVLA2026.01 | 85.2 | 83.3 | |
| State BackdoorVLA Model=OpenVLA2026.01 | 85.1 | 85.6 | |
| NoneVLA Model=OpenVLA2026.01 | 84.6 | — | |
| BadVLAVLA Model=OpenVLA2026.01 | 84.4 | 87.2 | |
| TrojanRobotVLA Model=π02026.01 | 84.3 | 86.3 | |
| TrojanRobotVLA Model=OpenVLA2026.01 | 84.1 | 84.8 | |
| BadVLAVLA Model=ACT2026.01 | 83.6 | 87.5 | |
| State BackdoorVLA Model=π02026.01 | 83.6 | 88.1 | |
| NoneVLA Model=DP2026.01 | 83.5 | — | |
| NoneVLA Model=SmolVLA2026.01 | 81.2 | — |