当前位置:网站首页>机器人强化学习——Learning Synergies between Pushing and Grasping with Self-supervised DRL (2018)
机器人强化学习——Learning Synergies between Pushing and Grasping with Self-supervised DRL (2018)
2022-07-04 23:11:00 【千羽QY】
论文地址:https://ieeexplore.ieee.org/document/8593986
1 简介
model-free 强化学习,Q-learning
方法:训练两个网络,分别预测像素级的push的Q-value 和 像素级的grasp的Q-value;Q-value最高的push或grasp被执行。
每个像素点的push被定义为从左向右推10cm;grasp被定义为以该点为中心,10cm为抓取宽度,水平抓取。
在测试时,图像被旋转16次,分别送入网络,因此可实现16个角度上的push和grasp。
本文采用高维action
,即抓取位姿和推;QT-Opt等方法采用更低维的action
,即末端偏移。
高维action
在全驱动系统
中是可行的,全驱动指物体的运动完全由机械手控制,如本文的抓取;
低维action
更适合于欠驱动系统
,需要根据系统反馈实时调整action,最终达到目标状态。欠驱动指物体的运动由环境和机械手同时决定,如预抓取、沿轨迹推物体等。
2 方法
state:RGB-D图像
action:在第一节描述
grasp reward:抓取成功 reward=1。如果机械手夹爪的张开长度大于阈值,则抓取成功。
push reward:场景图像的差值大于阈值 reward=0.5。该reward鼓励push动作使场景产生变化,但没有明确地使未来的抓取更方便。
Q网络结构:两个网络的结构相同。首先分别将RGB图像和D图像送入并行的DenseNet,然后合并特征,通过卷积和差值上采样输出预测Q-value。
1、如何给push设置reward
答:场景图像的差值大于阈值 reward=0.5。该reward鼓励push动作使场景产生变化,但没有明确地使未来的抓取更方便。
2、如何训练像素级预测网络
答:只对执行action的像素点p计算梯度,其他全为0
3 想法
1、本文方法本质上还是监督学习,只不过把grasp/push的置信度标签换成了reward,本质上一样
边栏推荐
- 实战模拟│JWT 登录认证
- Basic use and upgrade of Android native database
- Blue sky nh55 series notebook memory reading and writing speed is extremely slow, solution process record
- debug和release的区别
- PaddleOCR教程
- Actual combat simulation │ JWT login authentication
- heatmap. JS picture hotspot heat map plug-in
- ECCV 2022 | Tencent Youtu proposed disco: the effect of saving small models in self supervised learning
- Compare two vis in LabVIEW
- MP进阶操作: 时间操作, sql,querywapper,lambdaQueryWapper(条件构造器)快速筛选 枚举类
猜你喜欢
【kotlin】第三天
debug和release的区别
VIM editor knowledge summary
Why does infographic help your SEO
How long does it take to obtain a PMP certificate?
Redis: redis message publishing and subscription (understand)
[JS] - [sort related] - Notes
Actual combat simulation │ JWT login authentication
Compare two vis in LabVIEW
可观测|时序数据降采样在Prometheus实践复盘
随机推荐
uniapp 除了数字,其他输入无效
ECCV 2022 | Tencent Youtu proposed disco: the effect of saving small models in self supervised learning
The solution to the lack of pcntl extension under MAMP, fatal error: call to undefined function pcntl_ signal()
可观测|时序数据降采样在Prometheus实践复盘
A complete tutorial for getting started with redis: getting to know redis for the first time
头文件重复定义问题解决“C1014错误“
Redis introduction complete tutorial: List explanation
MariaDB's Galera cluster application scenario -- multi master and multi active databases
【kotlin】第三天
MP advanced operation: time operation, SQL, querywapper, lambdaquerywapper (condition constructor) quick filter enumeration class
QT personal learning summary
SPH中的粒子初始排列问题(两张图解决)
OSEK标准ISO_17356汇总介绍
The initial trial is the cross device model upgrade version of Ruijie switch (taking rg-s2952g-e as an example)
金融市场,资产管理与投资基金
实战模拟│JWT 登录认证
heatmap. JS picture hotspot heat map plug-in
Application of machine learning in housing price prediction
Selected cutting-edge technical articles of Bi Ren Academy of science and technology
MariaDB的Galera集群应用场景--数据库多主多活