|
Custom Algorithm pavan sathish from India [3 posts] |
9 year
|
hello sir, I am doing research in Artificial Intelligence and robotics area. I feel glad to find RR and wish to use it in my work. As a part of it, I want to create a custom algorithm. so please help me in this regard
|
|
|
Steven Gentner from United States [1446 posts] |
9 year
|
Pavan,
It will depend on how comfortable a programmer you are in which language. If you are ok with C/C++ then have a look at the plugins library where you can create a DLL plugin that runs at top speed and add that into RR. See the SwapColor module as an example:
http://www.roborealm.com/help/DLL_Program.php
but you can also use other languages that use different integration methods that while not as fast as a DLL do allow easier creation of modules:
http://www.roborealm.com/help/plugins.php
STeven.
|
|
|
pavan sathish from India [3 posts] |
9 year
|
hello steven sir,
That's nice of your information. Is the concept of "shape matching" comes under deep learning(machine learning)? since the object which is scanned is being identified under any orientation in shape matching function. please clarify me sir
|
|
|
Steven Gentner from United States [1446 posts] |
9 year
|
Pavan,
No, the shape matching is not considered deep learning. While deep learning is an active area of investigation and one of the tools we consider for processing images it is normally not the recommended approach. The reason is that things like shape matching is not a very difficult task that is not understood. The way basic shape matching works is to calculate one or more features (like the size of an object, or moment interia, or other descriptors) and compare those features with those features calculated in the template database. When features are similar in value then the shapes are similar in appearance. One could attach the results of these features into a deep learning network (as is done somewhat with a technique called Boosting) and train a more sophisticated classifier but for most of the uses of the shape matching a simple threshold works.
The other issue with Deep Learning is that one doesn't always know why a sample is classified in one way or another. In industrial applications one likes to know why things are classified the way they are to know when a system would fail. For production lines its important to know what and why a classification is wrong ... which can be difficult with Deep Learning.
STeven.
|
|
|
pavan sathish from India [3 posts] |
9 year
|
Dear STeven sir,
Thanks for giving me an idea on deep leaning. Actually as a part of my research I wished to select deep learning using RoboRealm tool. But now, after ur words I found it quite critical to dive into deep learning sir. 1) sir, will any of our roborealm modules support for deep learning? if so how and in what way can it be accomplished? 2)Scanning a particular object through camera/sensor in all views(3-D) and making a robotic gripper autonomous (i.e., dynamically opening and closing its fingers) to hold the object as per its size and shape without preloaded information , comes under Research?? can u better suggest me any possible way to do so or improvize my idea? I want to use our existing roborealm modules for that and some people they suggest me that as being a researcher in computer science Engg., you r suppose to devise an efficient algorithm rather than pre existing one. so please guide me sir taking out from this confusion.
regards
pavan
|
|
|
Steven Gentner from United States [1446 posts] |
9 year
|
Pavan,
If you are interested in exploring Deep Learning RoboRealm isn't the tool for that. We focus on getting projects done as quickly and efficiently as possible and thus look for the simplest way to solve the problem. Deep Learning is a great tool but not something we rely on and thus not something that is well represented in RoboRealm. You would be better off with another tool ... and, no, we don't know which one. I would start by typing in "Deep Learning Robot Gripper" into google and spend 3 weeks doing research into what other university projects have done in this field. There is a LOT of research in gripping objects and was even active in university back when I was majoring in robotics. I.e. you have a *lot* of reading to do. :-)
STeven.
|
|
|
Anonymous |
9 year
|
STeven sir,
Then I can jump into the research in "gripping objects"??
pavan.
|
|
|
Steven Gentner from United States [1446 posts] |
9 year
|
Yes, that may reveal interesting topics too. I'd also try "Deep Learning Robot Gripper" which I mentioned before. You may have to try a couple variations as to what keywords to use depending on what you see more in the literature. I.e. when you read something that is similar to what you want to do and they use a different phrase to describe it, search on that too!
STeven.
|
|
|
Anonymous |
9 year
|
STeven,
Thanks for your word that u would try on "Deep Learning Robot Gripper". In the mean while I ll also try. I ll wait eagerly for your valuable sources on the above topic
pavan
|
|