Event
MMGNet multimodal 6-DoF grasp synthesis paper released
1 reports · 1 sources
Rui Cao and other authors released a paper proposing MMGNet, a multimodal 6-DoF grasp synthesis network. Through hierarchical image-to-3D feature injection and rotational grasp degree modeling, it integrates multi-scale image features into 3D representations and jointly predicts grasp approach direction and in-plane rotation. The authors report that on GraspNet-1Billion using RealSense data, MMGNet achieves an average AP of 60.34, surpassing all comparison methods; in real-machine experiments on adversarial grasp objects, its grasp success rate is 8.46 percentage points higher than AnyGrasp. Ablation experiments show that the two modules bring complementary gains. The paper is currently under review, and the code, models, and data will be released after the paper is published.
Report timeline
- MMGNet: A Multimodal 6-DoF Grasp Synthesis Network for Cluttered Scenes
MMGNet proposes a multimodal grasp synthesis network that uses hierarchical image-to-3D feature injection and rotational graspness modeling to integrate multi-scale image features into 3D representations and jointly predict grasp approach direction and in-plane rotation. On GraspNet-1Billion with RealSense data, it achieves a mean AP of 60.34, surpassing all compared methods; in real-world experiments with adversarial grasp objects, its grasp success rate is 8.46 percentage points higher than AnyGrasp. Ablation studies show that the two modules provide complementary gains. Code, models, and data will be released after the paper is published.
arXiv cs.RO ·