Multi-Object Pick-and-Handoff 400 trials (simulation)
87.8Success RateSAM2Grasp
Evaluation results
| Method | Links | |
|---|---|---|
| SAM2GraspBackbone=SAM2 (Frozen), Strategy=Prompt-driven perception2025.12 | 87.8 | |
| ACT-CVAE-conditionInput Strategy=Visual Conditioning, Visual Cue=Target object bounding box rendered onto RGB2025.12 | 81 | |
| ACT-CVAEType=Original ACT model with CVAE, Level=policy-output level multimodality2025.12 | 50.8 | |
| ACTType=Standard deterministic Action Chunking Transformer2025.12 | 47.3 |