Hi community,
I am trying to record my own data for fine-tuning the OpenVLA model using a Trossen WidowX arm. I wonder what is the definition of state and action in the dataset?
I thought the state is the end-effector pose (x, y, z, r, p, y) + gripper state (0 for close, 1 for open) and the action would be (dx, dy, dz, dr, dp, dy, gripper) where (dx, dy, dz, dr, dp, dy) is the difference of the next state and the current state. But when I inspect the Bridge V2 dataset, it doesn't seem to be the case.
Thank you for your input!
Best regards,
Xiaowei
Hi community,
I am trying to record my own data for fine-tuning the OpenVLA model using a Trossen WidowX arm. I wonder what is the definition of state and action in the dataset?
I thought the state is the end-effector pose (x, y, z, r, p, y) + gripper state (0 for close, 1 for open) and the action would be (dx, dy, dz, dr, dp, dy, gripper) where (dx, dy, dz, dr, dp, dy) is the difference of the next state and the current state. But when I inspect the Bridge V2 dataset, it doesn't seem to be the case.
Thank you for your input!
Best regards,
Xiaowei