Handover Block on RoboTwin (Success Rate, Safe Rate)
0.75Success RateRDT-1B + PACT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RDT-1B + PACTPost-training=PACT, Fine-tuning Strategy=LoRA2026.06 | 0.75 | 0.82 | |
| π0.5 + PACTPost-training=PACT, Fine-tuning Strategy=LoRA2026.06 | 0.74 | 0.79 | |
| DP + PACTPost-training=PACT, Fine-tuning Strategy=Full-parameter2026.06 | 0.72 | 0.78 | |
| DP3 + PACTPost-training=PACT, Fine-tuning Strategy=Full-parameter2026.06 | 0.72 | 0.73 | |
| π0.5Post-training=None2026.06 | 0.66 | 0.64 | |
| RDT-1BPost-training=None2026.06 | 0.64 | 0.61 | |
| DP3Post-training=None2026.06 | 0.59 | 0.71 | |
| DPPost-training=None2026.06 | 0.28 | 0.7 |