🚀 Official Implementation of "PHYSICS-INFORMED GRASP DETECTION: ATTENTION-DRIVEN FORCE-CLOSURE ANALYSIS FOR DEXTEROUS MANIPULATION"
📜 Authors:Zimo Wen
🛠 Code: This repo provides the official implementation of our proposed grasp detection model.
Dexterous grasp detection is crucial for stable robotic manipulation, yet most methods rely on RGB-D or depth images, limiting their ability to utilize other robotic state inputs, such as grasp poses and joint configurations.
🔥 We propose DexGraspDetector, a novel approach integrating:
- Geometric Learning with PointNet++ for feature extraction
- Force-Closure Analysis for grasp stability
- An Attention Mechanism to balance multimodal inputs dynamically
📈 Our model achieves 93.2% accuracy, surpassing existing methods and demonstrating the advantages of integrating physical stability constraints.