Manipulating Point Prediction on Manipulating Point Prediction (Unseen)
16Success RateGRPO-MA
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPO-MATN=4, AN=42025.09 | 16 | |
| CoT-SFT2025.09 | 11.79 | |
| ManipLLM-7B2025.09 | 7.63 | |
| SFT2025.09 | 4.28 | |
| GRPOTN=4, AN=12025.09 | 3.94 | |
| GRPO-NoThinkTN=0, AN=162025.09 | 2.4 | |
| Qwen2.5-VL-3B2025.09 | 1.3 |