Utilizing Expert Data (.npz) in format of SB3.

[azafar1991]

Is it possible to use the trajectories (.npz) collected as expert response which is compatible for SB3 IRL models. ? I made an env which take user input to move an object in game. I collected the trajectories by providing user input , considering it to be expert response. But i am unable to find an example with which i can utilize that dataset.

[AdrianPrados]

I have the same problem, i have created a PyGame environment using also Gym to collect demonstrations by providing user input but i am not able to use that directly with the imitation implementation.

[Rajesh-Siraskar]

I too, plan to use a custom environment and generate trajectories for use as Expert data. It would be great to create a tutorial/example on these lines

[aPovidlo]

@azafar1991 @AdrianPrados @Rajesh-Siraskar Do you find the solution of the problem? Faced with a similar problem.

[Rajesh-Siraskar]

@azafar1991 @AdrianPrados @Rajesh-Siraskar Do you find the solution of the problem? Faced with a similar problem.

Hi Andrey, @aPovidlo :

This is what I have been able to achieve:

• Use of Imitation Inverse RL Libraries (as a demo integrated with Behavior Cloning algo.)
• Use of a custom environment that I built for predictive maintenance
• TensorBoard integration

[rizqisubeno]

Hi, Have you tried saving and loading with .npy format?. In my project, I successfully saved and re-loaded in every different scenario saving with different_name of *.npy file