[Question] Why the trained results of the example of Imitation Learning in the Isaac Lab are very bad?

[tkdggg]

Question

I collected 120 demonstrations for training, but the trained results are still very poor. Sometimes, the robotic arm moves towards the cube, but it consistently fails to grasp the cube.

[Mayankm96]

Hard for me to say. We had done experiments in the past with imitation learning and it worked fine. Although our scenario was simple (lifting a cube with a random initial position and fixed target).

This depends a lot on your setup (observations, actions) and the model you're using.

[tkdggg]

Thank you for your reply! Could you tell me how many demonstrations you collected in your training?

[Oliverbansk]

Your code is really awesome! I'm running into the same problem. Could you tell me what setup (observation space / action space) , collection technique (keyboard or SpaceMouse) and demo collection strategy (going freely or following some protocols like moving along xyz-axis in a specific order)? Thank you!