Hello! Thank you all for your great work and opensourcing pi. I am currently trying to understand the pi05 training with the droid dataset. However, I am seeing some conflicting information in the config file. Because of this I wanted to ask the following questions.
Is the config you used to train pi05_droid same as the pi05_full_droid_finetune config you provided in config.py? If so why is the action space and normalization stats different between pi05_full_droid_finetune and pi05_droid_finetune (ex. the action space for pi05_full_droid_finetune seems to be delta joint angles while pi05_droid_finetune is joint velocities)? If not would it be possible for you to provide the configs used to train pi05_droid?
Thank you for your time and assistance!
Hello! Thank you all for your great work and opensourcing pi. I am currently trying to understand the pi05 training with the droid dataset. However, I am seeing some conflicting information in the config file. Because of this I wanted to ask the following questions.
Is the config you used to train pi05_droid same as the pi05_full_droid_finetune config you provided in config.py? If so why is the action space and normalization stats different between pi05_full_droid_finetune and pi05_droid_finetune (ex. the action space for pi05_full_droid_finetune seems to be delta joint angles while pi05_droid_finetune is joint velocities)? If not would it be possible for you to provide the configs used to train pi05_droid?
Thank you for your time and assistance!