亿加合和智能车制作

标题: Some Ideas For Freescale Smart Car [打印本页]

作者: turfury00 时间: 2015-7-22 14:23
标题: Some Ideas For Freescale Smart Car
本帖最后由 turfury00 于 2015-7-22 14:56 编辑

Twice experience of Freescale Car National Competition,which has occupied half of my college period, is a really valuable memory for me.

Graduation one year ago.

Dropped out from my master university half years ago.

Now PhD.-ing.

It’s an amazing and struggling experience.

Now everything is OK and I finally catch some time to say something to my school brothers and my car friends or DaoYou.

Although I got a quite good prize,

I regret that I fail to achieve my idea in my car.

I regret that I have never made any contributions to the Freescale Forum (Actually, long time ago there were many good posts and ideas in the Forum, Now…Yeah…)

Sorry for “GuanShui”.

Sorry for typing in English now. (Because the school computer only supports English).

Maybe I can use PINGYING? , Oh Sh*t.

It has been a long time from I last time opened this forum website. And every year this time is really special for us. Exciting or …

I realized many interesting ideas, when I scanned over those posts in the Forum.

The post of Demon inspired me a lot, Zhuo is a great teacher who has been working on this competition for so many years.

Those ideas are some reasons which push me to say something here.

One “DaoYou” argues that the rules of this competition seem to be the constraints for us.

Good argument!

However, I just want to say rules are constraint for smart car, but not for our ideas which is the most important. Believe it or not, those ideas which can be achieved in our smart cars could be as good as or even better than techniques used by Google and other institutions.

作者: turfury00 时间: 2015-7-22 14:25
本帖最后由 turfury00 于 2015-7-22 14:38 编辑

“DaoYou” also mentioned Neural Network.

Yeah! Neural Network contains a lot good ideas, used in many real projects, have achieved excellent performance, which is also part of my research.

Human: Colourfulworld -> Brain -> decision

Machine: Information(010101011000) -> MCU -> decision

The information world of machine is always 01010100111--digital signal. So we need to write some programs and design some algorithms to let machines know the meaning of the digital signal and then make the decision.

The idea of Artificial Neural Networks (ANNs) is derived from Biological Neural Network; it is just designed to build a brain for the machine.

作者: turfury00 时间: 2015-7-22 14:27

[attach]79364[/attach]

In our brain, there are 60 billion neural cells which constitute the neural network and make us intelligent.

The picture above is one of structures of ANNs called single hidden layer feed-forward neural network. There are three layers: input layer, hidden layer, output layer.

The hidden layer always stands for the feature space of the input data.

作者: turfury00 时间: 2015-7-22 14:29
本帖最后由 turfury00 于 2015-7-22 14:42 编辑

Think in a mathematical way, ANNs set a large number ofneural nodes for the machine and combined with some mapping functions or sometimeswe call them weights.

Think in a pattern recognition way, ANNs first map the input data to the feature space (feature extraction), then separate out in the feature space and map to the output layer with suitable functions.

There are some relationships between the input and theoutput. ANN can be used to describe any relationships by function approximation. As for the smart car there are some relation between the roadinformation and the speed, direction which we want to control.

So the most important thing is the mapping function or the weights between the nodes. If we can train the weights in an optimal way, The ANNs can work really well.

So How to use and train the ANNs?

There are a lot of methods like Back-propagation.

Recently “Extreme Learning Machine” (ELM) is quite popular inthe machine learning area. If you have interests, you can search for some papers.

Many other methods like Optimal Weights Learning Machine (OWLM)are also very good.

Actually, a large number of ANNs based machine learningmethods have been proposed. Many of them have been achieve in some real projects. And the ideas of ANN can also be used in our smart car without doubt, ha-ha.

OK.

Maybe some other time, I will be happy to discuss ANNs with you guys.

And a question for all “Dao-You” (maybe someone get a goodprize or someone did not perform well), what is the most important thing that you received in this competition?

...

Enjoy something.

Cherish something.

Don’t be regret some years later.

All the best.

屌三年 22/07/2015

作者: L_X_ 时间: 2015-7-22 16:07

:o:o

作者: 歪腰 时间: 2015-7-22 16:11
醉了。。。。。。

作者: ①個亾◆◆潇灑 时间: 2015-7-22 17:33

:Q:Q:Q:Q

作者: 蜗牛会飞 时间: 2015-7-22 17:47

:@:@:@

作者: demon 时间: 2015-7-22 18:26
"The post of Demon inspired me a lot, Zhuo is a great teacher who has been working on this competition for so many years."
谢谢楼主的鼓励，论坛的学术气氛确实需要加强，现在满帖子都是6666，作为土鳖Ph.D Candidate表示很惭愧。

作者: A紫晶 时间: 2015-7-22 18:31
英语菜鸟一枚，坚持看完。很有感觉...

作者: 【征程】 时间: 2015-7-22 20:44

两次经验,飞思卡尔汽车国家竞争,占领了一半我的大学期间,对我来说是一个非常宝贵的记忆。
一年前毕业。我的主人大学辍学年前的一半。现在PhD.-ing。这是一个神奇和苦苦挣扎的经验。现在一切都好了,我终于抓住一些时间说一些我的学校朋友或DaoYou兄弟和我的车。
虽然我有一个不错的奖,我很遗憾,我不能实现我的想法在我的车。我很遗憾我从来没有做出任何贡献飞思卡尔论坛(事实上,很久以前有许多好的文章和想法的论坛,现在…是啊…)
对不起,“GuanShui”。对不起用英语打字了。(因为学校计算机只支持英语)。也许我可以用PINp拼音吗?哦shit。它已经很长一段时间从我最后一次打开这个论坛网站。每年这个时间对我们来说真的是特别的。令人兴奋或…我意识到许多有趣的想法,当我扫描那些帖子的论坛。
恶魔的职位我很多的启发,卓是一个伟大的老师一直致力于这方面竞争这么多年。
这些想法是一些原因把我插几句话。一个“DaoYou”认为,这种竞争的规则似乎对我们的约束。好论点!但是,我只想说规则是约束对智能车,但不是我们的想法是最重要的。不管你信不信,这些想法可以实现在我们的智能汽车可以一样好,甚至比谷歌和其他机构所使用的技术。

作者: 【征程】 时间: 2015-7-22 20:47
“DaoYou”也提到了神经网络。是啊!神经网络包含了很多好点子,在许多实际项目,取得了良好的性能,这也是我的研究的一部分。人类:Colourfulworld大脑- > - >的决定机:信息(010101011000)- >单片机- >的决定信息世界的机器总是01010100111——数字信号。所以我们需要编写一些程序和设计一些算法让机器知道数字信号的意义,然后作出决定。人工神经网络(ann)的想法是来自生物神经网络;它只是为了建立一个大脑的机器。

作者: 【征程】 时间: 2015-7-22 20:52
在我们的大脑中,有600亿个神经细胞构成的神经网络,让我们聪明。上面的图片是一个单隐层结构的神经网络前馈神经网络。有三层:输入层、隐层、输出层。隐层总是代表输入数据的特征空间。数学思想方法,人工神经网络设置大量ofneural节点机和结合一些映射函数或sometimeswe称之为权重。模式识别思想方法,人工神经网络输入数据映射到特征空间(特征提取),然后在特征空间分离出并与合适的函数映射到输出层。有一些输入和产量之间的关系。安可以用来描述任何关系函数近似。至于智能汽车有一些关系roadinformation和速度,我们要控制方向。所以最重要的事情是映射函数或节点之间的权重。如果我们能以最优的方式训练权重,人工神经网络可以工作得很好。那么如何使用和训练人工神经网络?有很多方法如神经网络。最近“极端学习机”(ELM)在机器学习领域中很受欢迎。如果你有兴趣,你可以搜索一些文件。像许多其他方法最优权重学习机器(OWLM)也很好。实际上,大量的基于人工神经网络的机器learningmethods已经提出。他们中的许多人已经在一些真正的实现项目。和安的想法也可以用于我们的智能汽车毫无疑问,哈哈。好吧。也许下一次吧,我会很高兴和你们讨论人工神经网络。
和一个问题“Dao-You”(也许有人goodprize或有人不执行),什么是最重要的事情在这个竞争,你收到了吗?…喜欢的东西。珍惜的东西。多年以后不要后悔。愿一切都好！
屌三年22/07/2015

作者: 六步上篮 时间: 2015-7-23 02:51
wow