Minor changes required to get the kuka repertoires to train and to visualize the kuka experiments

[ErickKramer]

Minor changes required to get the kuka repertoires to train and to visualize the kuka experiments

[riturajkaushik]

Hey, thanks for reminding. I forgot to correct these absolute paths. And yes, it seems env_params.txt were removed by gitignore. I'll anyways clean the codes a bit very soon and fix the issues.

[riturajkaushik]

I have updated the README to avoid the issue with the absolute path to the urdfs. Basically, you need to define an empty environment variable RESIBOTS_DIR before running the code. Script will automatically take the default paths to the urdfs when this environment variable is available. Regarding env_params.txt files, these files should not be required to run the examples.

Can you please confirm if you are able to run the mapelites script for the kuka experiment now?

[ErickKramer]

Hey, I cloned again your repo to test the comments you mentioned:

1. Creating the empty environment variable avoids the need for absolute paths
2. python3 kuka_pushing_exps/map_elites_kuka_pushing.py --toy 5 can be run without any problem.
3. python3 kuka_pushing_exps/kukaPushing_astar_ctlr2cartesian_v2.py --toy 0 --ucb_const 0.5 --kernel_var 0.003 --kernel_l 0.03 --visualization_speed 5.0 --search_size 800 --objectEulerAngles -1 --gui complains about the missing env_params.txt.

Traceback (most recent call last):
  File "kuka_pushing_exps/kukaPushing_astar_ctlr2cartesian_v2.py", line 487, in <module>
    main()
  File "kuka_pushing_exps/kukaPushing_astar_ctlr2cartesian_v2.py", line 289, in main
    with open(dd+"/env_params.txt", 'r') as f:
FileNotFoundError: [Errno 2] No such file or directory: './data/kuka_pusher_prior/kuka_pusher_map_toy_10_friction_1/env_params.txt'

[ErickKramer]

Hey @riturajkaushik, I have a question.
In

kaushik_2019_aprol/utils/map_elites.py

Line 45 in c577eb0

a = np.random.normal(0, (params["max"][i]-params["min"][i])/300.0, 1)

where does the 300.0 come from?. Or why did you decide for that value when computing the mutation?.

Thank you in advance!.

[riturajkaushik]

Hey, the 300.0 is actually a hyperparameter for the mutation variance. Since in the original CVT-map elites implementation it was kept like this, so I kept it as well. This value should not impact the optimization that much as long as it keeps the value of (params["max"][i]-params["min"][i])/300.0 small compared to the range of the parameters.

You can get the map-elites code from here if you want to use it for some other purpose.

Regarding the env_params.txt files, I'll upload them as soon as possible.

[riturajkaushik]

Since I lost the original env_params.txt files for the archives, I had to do a work-around to generate those files based on the information (e.g., friction, blocked legs, toy id etc.) stored in the directory names of the archives. Files are available now.