Apparently, this ongoing work is to make a preparation for futural research on Deep Reinforcement Learning. The goal of this work is to build a simulation platform that can insert the Deep Reinforcement Learning algorithms as a robot motion planning or navigation module.

I spent all day to position what I should do in this part of work. With an ambiguous goal and a big picture of the whole project, I am almost lost in the information I encountered in knowing about the related fields and knowledge.

In the end of the day, I hope what is now shaped in my mind is enough close to what my boss need. Below are what I believe to explain my work in the future.

Overview of the project

The project finally hopes to determine the robot's motion trajectory in real time by the method of Deep Reinforcement Learning. Given hundreds of different indoor environments the robot is trained to have the ability to move to a specified target without explicitly programmed navigation and obstacle avoidance algorithms.The following paper

The following paper are some related work that I believe could be useful.

  Target-driven visual navigation in indoor scenes using deep reinforcement learning

Active Object Localization with Deep Reinforcement Learning

Deep Neural Network for Real-Time Autonomous Indoor Navigation

Autonomous Navigation Planning with ROS

And a git book is quite complete in AI:

https://www.gitbook.com/book/leonardoaraujosantos/artificial-inteligence/details

ROS simulation

Gathering or building virtual indoor worlds

...

Replacing some modules in Navigation stack for Deep Reinforcement Learning algorithms

Navigation stack is a commonly used navigation module in ROS platform and SLAM tasks, of which the architecture is shown below. I believe some modules shown in this diagram can be replaced by DRL algorithm which I still need to dig deeper to determine.

A tutorial might walk through me all the way from building a customised robot to navigating this robot in Gazebo is offered here, which I would follow in the next few days and understand more on the relationship among those modules.

[Tutorial] Getting Starting with Autonomous Robots in ROS via Simulations

...

getting started with building a ROS simulation platform for Deep Reinforcement Learning的更多相关文章

  1. 中文译文:Minerva-一种可扩展的高效的深度学习训练平台(Minerva - A Scalable and Highly Efficient Training Platform for Deep Learning)

    Minerva:一个可扩展的高效的深度学习训练平台 zoerywzhou@gmail.com http://www.cnblogs.com/swje/ 作者:Zhouwan  2015-12-1 声明 ...

  2. Awesome Reinforcement Learning

    Awesome Reinforcement Learning A curated list of resources dedicated to reinforcement learning. We h ...

  3. AI佳作解读系列(三)——深度学习中的合成数据研究

    Below are some investigation resources for synthetic datasets: 1. Synthetic datasets vs. real images ...

  4. Evolutionary approaches towards AI: past, present, and future

    Evolutionary approaches towards AI: past, present, and future 2019-10-06 07:28:13 This blog is from: ...

  5. 斯坦福CS课程列表

    http://exploredegrees.stanford.edu/coursedescriptions/cs/ CS 101. Introduction to Computing Principl ...

  6. 【机器学习Machine Learning】资料大全

    昨天总结了深度学习的资料,今天把机器学习的资料也总结一下(友情提示:有些网站需要"科学上网"^_^) 推荐几本好书: 1.Pattern Recognition and Machi ...

  7. 【深度学习Deep Learning】资料大全

    最近在学深度学习相关的东西,在网上搜集到了一些不错的资料,现在汇总一下: Free Online Books  by Yoshua Bengio, Ian Goodfellow and Aaron C ...

  8. (转) Awesome Deep Learning

    Awesome Deep Learning  Table of Contents Free Online Books Courses Videos and Lectures Papers Tutori ...

  9. Top Deep Learning Projects in github

    Top Deep Learning Projects A list of popular github projects related to deep learning (ranked by sta ...

随机推荐

  1. spring命名空间不需要版本号

    为什么dubbo启动没有问题? 这篇blog源于一个疑问: 我们公司使了阿里的dubbo,但是阿里的开源网站http://code.alibabatech.com,挂掉有好几个月了,为什么我们的应用启 ...

  2. ectouch第七讲 之ECshop模板机制整理

    网上的资源感觉还是有些用,可以看看,帮助理解,ECshop模板机制整理原文:http://blog.sina.com.cn/s/blog_6900af430100nkn8.html 一.模板引擎: E ...

  3. 基本包装类型:Boolean、Number 和String

    /* var box = 'Mr. Lee'; //基本类型 //alert(box); //alert(typeof box); alert(box.substring(2)); //对象.方法(参 ...

  4. oracle nvl和nvl2的区别

    一直用oracle nvl函数,最近发现还有一个nvl2函数: nvl(a,b) 如果a不为null 则返回a,如果a为null则返回b; nvl2(a,b,c) ,如果a不为null 则返回b,如果 ...

  5. python :eval将字符串转换成字典

    #将字符串打印成字典 b=''' {'record': {'weight':20,'server':'100.1.7.9','maxconn':50},'backend': 'www.oldboy.o ...

  6. java静态块

    一般情况下,如果有些代码必须在项目启动的时候就执行的时候,需要使用静态代码块,这种代码是主动执行的 静态代码块的初始化顺序  class Parent{ static String name = &q ...

  7. Unity-Animator深入系列---录制与回放

    回到 Animator深入系列总目录 Animator自带了简单的动画录制,回放功能.但可惜的是不支持持久化的数据输出.因而不能作为录像保存 不过这种可以作为竞速,格斗类游戏在结束时经常出现的游戏回放 ...

  8. Linux如何查看当前占用CPU或内存最多的K个进程

    一.可以使用以下命令查使用内存最多的K个进程 方法1: ps -aux | sort -k4nr | head -K 如果是10个进程,K=10,如果是最高的三个,K=3 说明:ps -aux中(a指 ...

  9. Poj(3522),UVa(1395),枚举生成树

    题目链接:http://poj.org/problem?id=3522 Slim Span Time Limit: 5000MS   Memory Limit: 65536K Total Submis ...

  10. HDU(1572),最短路,DFS

    题目链接:http://acm.hdu.edu.cn/showproblem.php?pid=1572 很久没写深搜了,有点忘了. #include <iostream> #include ...