Robot Manipulation on Real-world Robot Tasks Pick Mug - Unseen
65Success RateRobotic Steering
Evaluation Results
| Method | Links | |
|---|---|---|
| Robotic SteeringTraining Strategy=Single, Training Task Setting=Single-Task Training (Place Marker in Mug only), Base VLA Model=π0.5, Number of demonstrations=20 per task, Number of trials=402025.11 | 65 | |
| LoRATraining Strategy=Single, Training Task Setting=Single-Task Training (Place Marker in Mug only), Base VLA Model=π0.5, Number of demonstrations=20 per task, Number of trials=402025.11 | 42.5 | |
| Robotic SteeringTraining Strategy=Joint Selection, Training Task Setting=Multi-Task Training (Place Marker + Place Cube), Base VLA Model=π0.5, Number of demonstrations=20 per task, Number of trials=402025.11 | 20 | |
| Robotic SteeringTraining Strategy=Non overlapping, Training Task Setting=Multi-Task Training (Place Marker + Place Cube), Base VLA Model=π0.5, Number of demonstrations=20 per task, Number of trials=402025.11 | 12.5 | |
| LoRATraining Strategy=Joint, Training Task Setting=Multi-Task Training (Place Marker + Place Cube), Base VLA Model=π0.5, Number of demonstrations=20 per task, Number of trials=402025.11 | 0 |