解释Roboschool Half-Cheetah的环境

时间:2018-02-03 15:27:05

标签: environment robotics reinforcement-learning openai-gym

我对roboschool Half-Cheetah有一些疑问。

  1. 我看到Half-Cheetah的观察空间是26.可以有人告诉我每个值是多少? - 我只计算了18.(同样,所有时间步长的某些值似乎都保持为0)

  2. 在roboschool / mujoco_assets下的half_cheetah.xml中,有以下注释:

    猎豹模型

  3. 状态空间按照它们的顺序填充关节 在此文件中定义。执行器也可以在关节上操作。

    State-Space (name/joint/parameter):
        - rootx     slider      position (m)
        - rootz     slider      position (m)
        - rooty     hinge       angle (rad)
        - bthigh    hinge       angle (rad)
        - bshin     hinge       angle (rad)
        - bfoot     hinge       angle (rad)
        - fthigh    hinge       angle (rad)
        - fshin     hinge       angle (rad)
        - ffoot     hinge       angle (rad)
        - rootx     slider      velocity (m/s)
        - rootz     slider      velocity (m/s)
        - rooty     hinge       angular velocity (rad/s)
        - bthigh    hinge       angular velocity (rad/s)
        - bshin     hinge       angular velocity (rad/s)
        - bfoot     hinge       angular velocity (rad/s)
        - fthigh    hinge       angular velocity (rad/s)
        - fshin     hinge       angular velocity (rad/s)
        - ffoot     hinge       angular velocity (rad/s)
    
    
    Actuators (name/actuator/parameter):
        - bthigh    hinge       torque (N m)
        - bshin     hinge       torque (N m)
        - bfoot     hinge       torque (N m)
        - fthigh    hinge       torque (N m)
        - fshin     hinge       torque (N m)
        - ffoot     hinge       torque (N m)
    

    如果此处提供的订单与他们在观察表中出现的顺序相同,请您确认一下吗?如果是这样,我应该考虑始终为0的值吗?

    谢谢。

0 个答案:

没有答案