Vision-language Robotic Manipulation on RLBench simulation (test)
80Success Rate (Cup)RADAR
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| RADARRollouts per task=10, Demonstration setting=1-shot, Initial state randomization=True, Environment reset protocol=Simulation ground truth reset, Skill retrieval mechanism=VLM2026.03 | 80 | 80 | 90 | 100 | 80 | 100 | 70 | 80 | 40 | 90 | |
| ReKepRollouts per task=10, Demonstration setting=1-shot, Initial state randomization=True, Environment reset protocol=Simulation ground truth reset2026.03 | 20 | 10 | 10 | 40 | 40 | 40 | 20 | 10 | 0 | 20 | |
| MOKARollouts per task=10, Demonstration setting=1-shot, Initial state randomization=True, Environment reset protocol=Simulation ground truth reset2026.03 | 20 | 30 | 10 | 40 | 10 | 30 | 20 | 0 | 0 | 10 |