xbpeng / DeepMimic

Motion imitation with deep reinforcement learning.
https://xbpeng.github.io/projects/DeepMimic/index.html
MIT License
2.3k stars 486 forks source link

How can I train a new policy with a task metioned in the paper, like throw? #100

Open hellozjj opened 4 years ago

hellozjj commented 4 years ago

hello: I notice that the all configurations in the args file are not include the tasks metioned in the paper. How can I train a new policy with a task (how should I configure/wirte the arg file?). Another question is about the environments. The paper says you have four environments, but there is only one environments (plane) in your code or just I misunderstanding something?

hellozjj commented 4 years ago

And there are also not Multi-Skill Integration in the code?

hellozjj commented 4 years ago

Is this code only a part of the original code?

xbpeng commented 4 years ago

This code is only for training policies for motion imitation. It does not include the other tasks.

BenJaminB1ue commented 4 years ago

Hello Peng @xbpeng Could you tell me how can we get the whole code, like spinkick or throw to the goal target, please? I'm really long for that.

Zju-George commented 4 years ago

Hello Peng @xbpeng Could you tell me how can we get the whole code, like spinkick or throw to the goal target, please? I'm really long for that.

I guess they won't put the whole code out... But I am going to realize the physical based animation in Unity and train them using unity ml-agents in a short future.

BenJaminB1ue commented 4 years ago

Hello Peng @xbpeng Could you tell me how can we get the whole code, like spinkick or throw to the goal target, please? I'm really long for that.

I guess they won't put the whole code out... But I am going to realize the physical based animation in Unity and train them using unity ml-agents in a short future.

你好,George! 这也正是我近期的研究内容,您也可以参考这一下这些工作: https://github.com/Sohojoe/ActiveRagdollStyleTransfer https://github.com/tanakataiki/YAD2U 您愿意提供学习或交流的平台吗? BTW我的邮箱yihewu@njust.edu.cn

Zju-George commented 4 years ago

Hello Peng @xbpeng Could you tell me how can we get the whole code, like spinkick or throw to the goal target, please? I'm really long for that.

I guess they won't put the whole code out... But I am going to realize the physical based animation in Unity and train them using unity ml-agents in a short future.

你好,George! 这也正是我近期的研究内容,您也可以参考这一下这些工作: https://github.com/Sohojoe/ActiveRagdollStyleTransfer https://github.com/tanakataiki/YAD2U 您愿意提供学习或交流的平台吗? BTW我的邮箱yihewu@njust.edu.cn

感谢您,原来已经有人做啦!我还不知道呢! 学习和交流的平台指?我如果成功做出来可能会在b站做视频,你可以在我的github个人简介里找到!