search

xbpeng / DeepMimic

Motion imitation with deep reinforcement learning.
https://xbpeng.github.io/projects/DeepMimic/index.html
MIT License
2.31k stars 488 forks source link

Can we ran DeepMimic directly in DeepMimicCore's main.cpp? #70

Open Zju-George opened 5 years ago

Zju-George commented 5 years ago

@xbpeng @wangshub @0xflotus @376TEMP Also, if I want to debug and change the code, I must change C++ code?

I wonder, if I want to write the pose data out into a json file when glutMainloop is running, should I change the C++ code and ran it directly in main.cpp in DeepMimicCore?

And after debugging in C++ code, rewrap it using SWIG?

I'm a noob in both C++ and python, please give me some advice.

ZhengyiLuo commented 5 years ago

Yes, you need to change the behavior of the bullet environment you need to change C++ code and recompile. If you are using windows in the sln there should be a python target you can build the python wrapper.

You can not run the policies in main.cpp I think, since the policies can only be run in python using tensorflow.

Also, please stop spamming other threads when you have build questions :). Specifically: This one We all have questions and we can help each other as much as possible, but don't spam since the authors have limited time and we should appreciate their help.

xbpeng commented 5 years ago

That's right, you can edit the C++ code in DeepMimicCore and debug in C++. Then when you are ready you can run SWIG to build the python wrappers and train using the python framework.

You cannot run the policies in C++ because all of the RL framework is in python.

Zju-George commented 5 years ago

yes we can do that, but it seems that we can't get the trained model run properly, but as for running mocap data, it's fine. I will dig into more, thx anyway. image

Zju-George commented 5 years ago

@xbpeng Thanks for your reply! Could you please share some more ground types cuz there is only "dull" plane.. image