How you get the controlling data used for trainning?

[milong26]

I mean, you said predict $\hat{y}$ then training the model to get the best $\hat{y}$ which is similar to $y$. how u get the $y$? Adjusting by mujoco by yourself?

Thanks!

[kochlisGit]

Check the script: generate_expert_dataset.py

[milong26]

Check the script: generate_expert_dataset.py

Thank you for replying! I read from generate_expert_dataset, getting that transition_history = simulator.transition_history. then the transition can be written to 'data/expert_dataset.csv'.
I can't understand how the simulator generate correct control data ;-;

[kochlisGit]

Well, it calculates several midpoints between an initial and a final value of each actuator. These values are manually defined by the user, depending on the trajectory that they wish to implement.