ee pose vs joint pose

[DelinQu]

Hi, In the pi0 paper, it is mentioned that the state information during model training is joint poses. The tuning examples provided in this repository also use joint pose, except for libero. I suspect that the libero simulator prefers to provide EE pose. I would like to ask why joint pose is prioritized for model training instead of EE pose? What are the experiences or considerations behind this choice? 🤔