zoukankan      html  css  js  c++  java
  • Deep RL Bootcamp Lecture 4A: Policy Gradients

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

    in policy gradient, "a" is replaced by "u" usually.

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     use this new form to estimate how good the update is.

     

     

     

     

     

     

     

     

    If all three path show positive reward, should the policy increase the posibility of all the sampling?

     

      

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     monte carlo estimate

     

     

     

     

     

    TD estimate

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

     

    2 weeks to train as respect to real world time scale.

    but could be faster in emulator (MOJOCO).  

    we don't know whether a set of hyperparameter is going to work until enough interations have past. So it's kind of tricky, and using emulator could alleviate this problem.

     

     

    question: how to transform learnt knowledge of robot to real life if we are not sure about the match between simulator and real world?

    Randomly initilize many simulator and see the robustness of the algorithm

     

     

     

     

     

     

    this video shows that even a robot with two years of endeavor of a group of experts still isn't good at locomotion

    hindsight experience replay

    Marcin Richard from OpenAI

    the program is set to find the best way to get pizza, but when the agent find a ice cream, the agent realizes that ice cream, corresponding to a higher reward, is the exact thing it wants to get. 

     

     

     

     https://zhuanlan.zhihu.com/p/29486661

     https://zhuanlan.zhihu.com/p/31527085

  • 相关阅读:
    Spring核心思想:IOC(控制反转)、DI(依赖注入)和AOP(面向切面编程)
    synchronized 与 lock锁的异同
    SpringMVC工作流程
    close()和flush()的区别
    MySQL—索引(Index)
    MySQL—事务(ACID)
    MySQL—存储引擎
    周学习笔记(16)——大三下
    周学习笔记(15)——大三下
    周学习笔记(14)——大三下
  • 原文地址:https://www.cnblogs.com/ecoflex/p/8974602.html
Copyright © 2011-2022 走看看