update_specalist not a function of environment MikeStepperEnv-v0.

[fabianbunbury]

very excied to use this code to see the demo from https://www.youtube.com/watch?v=Qe1izKvpFRQ using the command:
python3 enjoy.py --env mocca_envs:MikeStepperEnv-v0 --net models/mocca_envs:MikeStepperEnv-v0_latest.pt
but i get an error telling me " mike_StepperEnv object has no attribute 'update_specialist''.

If I delete env.update_specialist() I get further errors related to a missmatch in the size of the environment envs and the expected input size for the nn-model.

help resolving this problem would be much apreciated as I am very excited about trying out this excelent work!!!

[belinghy]

I think it might have been caused by a version mismatch between this repo and the mocca_envs module. We've updated the master branch to make demos with pretrained controllers work. Please see the new instructions and let us know if there are still any issues with running the code.

Thank you for trying out our work!

[fabianbunbury]

Thank you very much Ben I have sorted the problem and it is all working brilliantly! you are right it was an issue with using the wrong branches. i also had an issue with torch-- i needed to go back a version to 1.5.. have you ever tried running your code with quadrupeds? i have been amazed by the quadruped locomotion learning work that has been done in pybullet, but think an elegant quadruped stair climber would be brilliant.

…

On Mon, Sep 14, 2020 at 9:42 PM Ben Ling ***@***.***> wrote: I think it might have been caused by a version mismatch between this repo and the mocca_envs module. We've updated the master branch to make demos with pretrained controllers work. Please see the new instructions and let us know if there are still any issues with running the code. Thank you for trying out our work! — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC2Y73PMKVFRJGDBGGUUROTSFZ525ANCNFSM4RLMKREQ> .

-- Fabian Bunbury Engineer at Dyson. 07572041123

[belinghy]

For the pytorch issue, the change to reset_dist in this commit should be compatible with the newest pytorch version.

I have tried (very briefly) quadruped stepping stone, but it needs more work to get it working. I think first training a controller with reference motion, then adapting the controller to stepping stone should be easier than training from scratch.