Skip to content

Imitation Learning Example

This entry applies to DR02 Pro. deep-robotics-mimic uses Isaac Lab to train and evaluate policies that track reference motions.

Data and training workflow

Section titled “Data and training workflow”
  1. Obtain robot reference motions through the motion-retargeting example, or use a format supported by the repository.
  2. Follow the current repository guide to prepare Isaac Lab and convert reference motions to forward-kinematics NPZ training data. Do not assume arbitrary NPZ files share the required fields or coordinate conventions.
  3. Inspect motions with the replay tools before choosing the training task and reference dataset. Use the current README for environment, dependencies and training parameters.
  4. Replay and evaluate the trained policy. After simulation evaluation, check deployment requirements against the repository instructions and the matching DR02 Pro SDK workflow. Completing training does not establish hardware validation.
Section titled “Related resources”