train

[Nimingez]

For each timestep: observations

• images
  • cam_high (480, 640, 3) 'uint8'
  • cam_low (480, 640, 3) 'uint8'
  • cam_left_wrist (480, 640, 3) 'uint8'
  • cam_right_wrist (480, 640, 3) 'uint8'
• qpos (14,) 'float64'

• qvel (14,) 'float64'

  action (14,) 'float64' """

  data_dict = { '/observations/qpos': [], '/observations/qvel': [], '/observations/effort': [], '/action': [], } May I ask if qvel and effort are used in the training phase? What about in the inference phase? What would happen if qvel and effort are missing during the training phase?

[tonyzhaozh]

Hi,

qvel and effort are logged but we haven't tried including them in training/inference. Missing these fields should not matter at all.

Tony