When using the Allegro version of the code to train other hands such as LinkerHand, I discovered that the results of the cal_distance and cal_self_distance functions in the handmodel were incorrect.

[yys-abu]

When using the Allegro version of the code to train other hands such as LinkerHand, I discovered that the results of the cal_distance and cal_self_distance functions in the handmodel were incorrect. At the outset of training, even when the hand is far from the object, one might still claim significant penetration.

[djqqqqqqqqqqq]

Have you solved the problem?I modified generate_grasps.py from shadowhand version to train my linkerhand and it works out fine.