Robotic Manipulation on Physical-world environment UR3e manipulator
89Completion Accuracy (CA)Clean baseline
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Clean baselineRBA scheme=w/o, VLM=Qwen-vl-max [77]2024.11 | 89 | — | |
| TrojanRobot (Prime-S)RBA scheme=Ours (Prime-S), VLM=Qwen-vl-max [77]2024.11 | 89 | 72 | |
| TrojanRobot (Prime-I)RBA scheme=Ours (Prime-I), VLM=Qwen-vl-max [77]2024.11 | 89 | 83 | |
| Clean baselineRBA scheme=w/o, VLM=Qwen-vl max-latest [77]2024.11 | 80 | — | |
| TrojanRobot (Vanilla scheme)RBA scheme=Vanilla scheme, VLM=Qwen-vl-max [77]2024.11 | 80 | 19 | |
| TrojanRobot (Prime-S)RBA scheme=Ours (Prime-S), VLM=Qwen-vl max-latest [77]2024.11 | 80 | 74 | |
| TrojanRobot (Prime-I)RBA scheme=Ours (Prime-I), VLM=Qwen-vl max-latest [77]2024.11 | 80 | 76 | |
| TrojanRobot (Vanilla scheme)RBA scheme=Vanilla scheme, VLM=Qwen-vl max-latest [77]2024.11 | 72 | 24 | |
| TrojanRobot (Prime-P)RBA scheme=Ours (Prime-P), VLM=Qwen-vl-max [77]2024.11 | 72 | 50 | |
| TrojanRobot (Prime-P)RBA scheme=Ours (Prime-P), VLM=Qwen-vl max-latest [77]2024.11 | 69 | 48 | |
| Clean baselineRBA scheme=w/o, VLM=OWLv2 [42]2024.11 | 35 | — | |
| TrojanRobot (Prime-S)RBA scheme=Ours (Prime-S), VLM=OWLv2 [42]2024.11 | 35 | 33 | |
| TrojanRobot (Prime-I)RBA scheme=Ours (Prime-I), VLM=OWLv2 [42]2024.11 | 35 | 19 | |
| TrojanRobot (Prime-P)RBA scheme=Ours (Prime-P), VLM=OWLv2 [42]2024.11 | 33 | 17 | |
| Clean baselineRBA scheme=w/o, VLM=MiniGPT-v2 [4]2024.11 | 31 | — | |
| TrojanRobot (Vanilla scheme)RBA scheme=Vanilla scheme, VLM=MiniGPT-v2 [4]2024.11 | 31 | 9 | |
| TrojanRobot (Prime-S)RBA scheme=Ours (Prime-S), VLM=MiniGPT-v2 [4]2024.11 | 31 | 43 | |
| TrojanRobot (Prime-I)RBA scheme=Ours (Prime-I), VLM=MiniGPT-v2 [4]2024.11 | 31 | 0 | |
| TrojanRobot (Vanilla scheme)RBA scheme=Vanilla scheme, VLM=OWLv2 [42]2024.11 | 30 | 15 | |
| TrojanRobot (Prime-P)RBA scheme=Ours (Prime-P), VLM=MiniGPT-v2 [4]2024.11 | 26 | 24 |