反向运动学 [英] Inverse Kinematics

问题描述

解决机器人逆运动学的最佳方法是什么？神经网络（模糊逻辑）还是分析？你有神经网络解决的逆运动学源代码吗？谢谢。



我已经用分析方式实现了它。最近我读了一些文章，说明神经网络在解决反向运动学方面更有效。我需要一个源代码，因为我没有时间修改我的工作了。我不太了解神经网络，我相信它将极大地帮助我的项目提高效率。如果我问得太多，我很抱歉。

What is the best way to solve inverse kinematics of a robot? Neural network (fuzzy logic) or analytical? Do you have a source code of inverse kinematics solved by neural network? Thank you.

I have already implemented it using analytical way. Recently I have read papers stating that neural network is more efficient in solving inverse kinematics. I need a source code since, I do not have time anymore to revise my work. I do not understand neural networks that much and I believe it will greatly help my project for it to be efficient. I''m sorry if I am asking too much.

推荐答案

你的问题看起来很奇怪;所以我觉得有点困惑。我对机器人运动学问题的第一反应是：它甚至不是分析性的，它是小学的几何学！如果你考虑一下我工作的简单工业机器人，就会出现这种情况：一个平原上的2-3个接头加上一个...我可以看到，这是多数。但是，我明白，可能有复杂的系统;我知道神经网络的使用，但那是真正复杂的系统和实际存在的不适定问题。



我不知道理解，为什么你在问我已经用分析方法实现它时，问为什么是神经网络方法。你已经做到了并且做得正确，这是你问题简单的一个指标。在这种情况下，你怎么能想到神经网络方法的适用性？而且，如果你不了解你的机器人问题和机器人的复杂性，你怎么能希望得到任何明确的答案？我有一种不愉快的感觉，你自己很难理解这件事。如果你证明我错了，我会很高兴的。对我而言，分析解决方案的存在似乎使其他一切变得荒谬。做什么的？只是为了时髦？



我也无法理解在神经网络（模糊逻辑）中使用括号。你在谈论所谓的神经模糊方法吗？通常，神经网络和模糊逻辑（模糊集和所有这些）是不同的数学分支;我对神经网络很不熟悉，但模糊集和逻辑是相对容易学习的东西。我再次对你的问题感到困惑。如果在你的问题中有可能使用这种高级形式主义，那么无论如何你都会学到这些东西，但如果你熟悉背景，那么你的问题会有所不同。你真的明白你在问什么吗？同样，我感觉你的问题不是技术问题，而是整个有条理的知识及其应用方法。如果你证明我错了，我会很高兴的。



我在神经网络，遗传学和机器人模糊方法方面的最佳专家是Anorge Kirillov，AForge.NET Framework的作者：AForge.NET开源框架 [ ^ ]， http://en.wikipedia.org /wiki/AForge.NET [ ^ ] 。他是CodeProject的作者，您可以通过CodeProject直接与他联系。



我在CodeProject上看到了其他或多或少的相关作品，但您需要执行搜索你自己。

-SA

Your question looks very strange; so I feel a bit confused. My first reaction of the question on robot kinematics was: it''s not even "analytical", it''s elementary school geometry! And this would be true if you consider simple industrial robots I worked with: 2-3 joints in one plain plus one more... As I could see, this is a majority. However, I do understand, there can be complex system; and I know about the use of neural networks, but that''s for really complex systems and ill-posed problems that actually exist.

What I don''t understand, is why are you asking about neural network approach while saying "I have already implemented it using analytical way". It you already did it and did correctly, this is an indicator of simplicity of your problem. How can you even think of applicability of neural network approach in this case? Moreover, how can you hope for any definitive answer if you conduct no idea of your robot problem and the complexity of your robot? I have an unpleasant feeling that you poorly understand this matter yourself. I would be very happy if you prove me wrong. To me, it looks like the presence of analytical solutions makes everything else an absurd. What for? Just to be "fashionable"?

I also cannot understand using brackets in "Neural network (fuzzy logic)". Are you talking about so called "neuro-fuzzy" approach. Normally, neural network and fuzzy logic (fuzzy sets and all that) are different branches of mathematics; I''m poorly familiar with neural networks, but fuzzy sets and logic is relatively easy stuff to learn. Again, I''m confused with your question. If there is a potential of using such advanced formalism in your problem, you would learn this stuff anyway, but if your were familiar with the backgrounds, your questions would be different. Do you really understand what are you asking about. Again, I have a feeling your problem is not a technical one but a whole methodical approach to knowledge and its application. And again, I would be very happy if you prove me wrong.

Best specialist I know in neural network, genetic and perhaps fuzzy methods in robotics is Anrew Kirillov, author of AForge.NET Framework: AForge.NET open source framework[^], http://en.wikipedia.org/wiki/AForge.NET[^]. He is a CodeProject author, you may want to contact him directly through CodeProject.

I saw other more or less relevant works on CodeProject, but you need to perform your search yourself.

—SA

我的两分钱：

My two cents on:

如果有分析解决方案，那么它将胜过神经网络一个。

"if there is an analytical solution then it will outperform the neural network one".

：）

:)

我见过很多DOF的最佳解决方案和有限的开发预算是利用可用的廉价游戏物理引擎。 ''Pin'''''''''''''''''''''''''''''''' ''挑选'一个末端效应器并移动它。您将看到正在计算的关节位置。所有繁重的工作都完成了。有伪代码和更多信息此处。

The best solution I''ve seen for many DOF and a limited development budget is to take advantage of the cheap game physics engines available. ''Pin'' the body of a ''ragdoll''. ''Pick'' an end effector and move it. You will see joint positions being calculated. All of the heavy lifting is done. There is pseudo code and more info here.

这篇关于反向运动学的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！