search

tengyu-liu / GenDexGrasp

Code Repository for GenDexGrasp: Generalizing Dexterous Grasping across Robotic Hands via Contact Map Matching
93 stars 10 forks source link

Fig7:GenDexGrasp in tabletop scenarios #11

Open tjwyj opened 1 year ago

tjwyj commented 1 year ago

I want to know what's the proper training? Thanks for your answer!

Xiaoyao-Li commented 1 year ago

Thanks for your interest.

In GenDexGrasp, we test a tabletop grasping scenario in simulation. First, we employ DFC to synthesize feasible grasping poses for tabletop scenario. This is achieved by introducing a penetration penalty between the robotic hand and the table. Then, we utilize the synthesized MultiDex Dataset, focused on tabletop scenarios, to re-train the ContactMapCVAE network. We caputure the point cloud and estimate its normals for the object and then generate grasping pose with generated contact map given by the ContactMapCVAE.

tjwyj commented 1 year ago

Thanks for your answer. Could you public the tabletop scenario related datasets if it's convenient for you？Or share it for a short time? Best regards,