How to train and evaluate algorithms on the robot_env?

[007lbx]

1. I follow the scripts file to train the algorithms, but I want to known where is the use of the train_new.py ?
2. When I use the eval.py to evaluate the algorithms, there are some problems:
   Traceback (most recent call last):
   File "src/eval.py", line 117, in
   main(args)
   File "src/eval.py", line 55, in main
   action_repeat=args.action_repeat,
   AttributeError: 'Namespace' object has no attribute 'action_repeat'
   I find this eval.py is same to the \DMC\src\eval.py, but it's obvious that these two issues are different.

[DavidBert]

Hey!
Sorry these two files should not be included in the public repo, its a mistake.
The train_new.py file was just a sand bow we used to get random pictures of the environment. If you look closely it's not a training loop but just a script that performs a random trajectory and save an image of the last state.
The eval file is just a plain copy of the one in \DMC without any modification. It is just a remnant files that should not be present in this folder. If you want an equivalent evaluate.pyfile for the robot_env you should take inspiration from the one in \DMCand modify the evaluatemethod to match with the one implemented here.
Sorry for this oversight on our part to delete these files. It's now done, thank you for pointing this out to us.

[007lbx]

Thanks for your replay!
The appendix D provides additional results on a vision-based robotic environment:" We modified the original simulator to include three testing environments for both tasks......". The "Test 1-3"is equivalent to the Robotic Manipulation introduced by
(Jangir et al. 2022) ? Can I adopt four test environments "Test 1-4" for each task?

[DavidBert]

It's been a while since I did not interacted with it but yes I think so.