I have seen loads of robots that are controlled via the internet, but I wonder if anyone has ever made a robot that actually gathers information from the internet?
Such as...Looking at an object, and then by you "tagging" certain sites, like Photobucket or Dictionary.com, it would find a picture that looks relatively similar to the object it found, and then it would "read" the keywords to see what that object is. (So it would say "Oh. The majority of people on the website say that this object is a tissue box, so I am seeing a tissue box.)
Either that, or a separate server of information you can edit so that the robot can still look up information only in that server?
Not only possible but it has been done wiht Cameras and stationary computers. The biggest challenge you would have with this in a robot is the mobile computing platform for the robot is going to have to be VERY powerfull. Since alot of Micro controllers live in the 8 - 75 MHZ range with less then a Meg of memory (most ARM processors have more memory), and the computers I have seen do this were dual core - dual processor beasts running between 5-6 GHZ with between 6-10 gig of ram (Massive overkill, but some where in between is the actual requirement)
However if you were to take say a 2.4 GHZ pc (old P4) with a gig of ram and use it as the brain for the robot, then it just comes down to programming.
Image processing for feature Image processing for feature recognition is a pretty intensive task - you’d need something pretty powerful to handle that on top of all the other data transfer and connectivity functions. Most hobbyist level microcontrollers would fall well short of this task, unless you intend to have a PC do most of the grunt work and transmit the results back to the bot.
Check out the OpenCV pages Check out the OpenCV pages in this product manual on ViewPort. I have not this feature yet on my robot but I am toying with the idea
Start writing the network software Google will probably have service “find images like this one” available by the time we settled on thinking it can’t be done…
Last year I wrote a visual search engine that functions in exactly this way: you supply it with an image, and it scours the database for the best matches according to various visual markers. The program was written in C++ using OpenCV (no surprises there =D).
A few years ago there was a software team that had an online internet imaging search engine, I think they were French, but sadly it is no longer available. Their program was much like mine, only it could search the internet, and it was probably a lot more accurate. No ‘tags’ are used in the search process, only the visual image data itself.
As it required the attention of a whole PC to run (and took several seconds to return the search results, even with a database of only a few hundred) I felt it was not suitable for use in a real-time robotics environment. Having said that, it was rated one of the best programs in the class, so that gives you an idea of how intensive this kind of analysis can be.
I don’t have the source code on my computer at the moment, but I’ll see if I can drag it off the university servers. This page contains a lot of the material we were taught as part of the paper this project was for.
