CS294-112深度增强学习课程（加州大学伯克利分校 2017）NO.4 Learning policies by imitating optimal controllers

 

 

 

 

 

 

 

There are some problems: mismatch of model and reality; gradient explosion

 

 

 

 

 

 

 

 

 

 

 

 

so, the dynamics can be quite messy, and backpropogating can be quite problematic.

 sudden change in velocity and so on. schochastic system. gradient descent can be tough.

 

 

 

 

can we apply this trajectory optimization method to optimize policy?

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

GPS: guided policy search

 

 

 

 

in this case, o_t is from the camera and the joint velocity

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

                     

 

                       

 

                       

 

 

 

 

                 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

https://katefvision.github.io/katefSlides/imitate_controlers_katef.pdf