Slashdot Mirror

I frequently have to create large collections of images from all sorts of file types -- some text-based, some graphics -- that get housed in a collection of images for easy, standardized review. If there were something that could avoid the step of extracting text from them, or later OCRing them and still end up with a searchable image collection, well, that would be exceedingly cool. It would cut the initial time outlay I have to devote to virtually any given project I have to deal with by 25 to 50%.

Have you considered trying some SIFT-based algorithms? It usually performs pretty well on detecting matches between 2D forms, if I'm understanding your problem correctly. There's some open-source libraries that implement SIFT, as well as a pretty nice demo from Evolution Robotics that applies SIFT to data from USB cameras, so you can have it do things like recognize dollar bills, book pages, etc.

WowWee and Evolution robotics announced a partnership last March which will probably get Evolution's object recognition technology onboard the Robosapien. This is the same technology that the AIBO used to recognize its special command cards. WowWee might have to upgrade Robosapien's processor for this because the image processing for the object recognition is expensive to do. However, I see that Evolution has an optimized version of the software for mobile phones so that might help a lot. Expect the next Robosapien version to be more capable but also a bit more expensive.

Hm... I originally included a couple of links which show how the technology works, but they were unfortunately removed from the submission. (Also, in retrospect I really should have made the title "Next-gen Robot Toys Smart Enough to Fetch Beer," as it probably would have increased the level of discussion a little more.) I'll go ahead and add them here...

The first bit of technology Evolution Robotics will probably be contributing is their ViPR (Visual Pattern Recognition) tech, which allows for real-time recognition of objects in the environment. It's really quite impressive to see it in action -- it can learn how an object looks using just a single training example, has a high recognition rate, is resilient to occlusion/rotation/scale, and can operate at 15fps on an ordinary computer. It works by efficiently extracting a few hundred SIFT (scale-invariant feature transform) features from an image, and then learns what affine arrangement of them indicate an object. A downloadable demo is available on the ViPR page.

Their SIFT-based ViPR techniques work -very- well on rigid objects, like labels, signs, furniture, cups, and pretty much any sort of static pattern. They tend to work less well on deformable objects, like faces and people.

The other piece of technology is NorthStar, used for indoor navigation. This uses a projector to shine some IR light spots on the ceiling. An IR-sensitive camera on the robot can see where the light spots are, allowing it to easily determine its location relative to the projector, so it can basically create a map of its environment using projector-relative coordinates.

Hm... I originally included a couple of links which show how the technology works, but they were unfortunately removed from the submission. (Also, in retrospect I really should have made the title "Next-gen Robot Toys Smart Enough to Fetch Beer," as it probably would have increased the level of discussion a little more.) I'll go ahead and add them here...

The first bit of technology Evolution Robotics will probably be contributing is their ViPR (Visual Pattern Recognition) tech, which allows for real-time recognition of objects in the environment. It's really quite impressive to see it in action -- it can learn how an object looks using just a single training example, has a high recognition rate, is resilient to occlusion/rotation/scale, and can operate at 15fps on an ordinary computer. It works by efficiently extracting a few hundred SIFT (scale-invariant feature transform) features from an image, and then learns what affine arrangement of them indicate an object. A downloadable demo is available on the ViPR page.

Their SIFT-based ViPR techniques work -very- well on rigid objects, like labels, signs, furniture, cups, and pretty much any sort of static pattern. They tend to work less well on deformable objects, like faces and people.

The other piece of technology is NorthStar, used for indoor navigation. This uses a projector to shine some IR light spots on the ceiling. An IR-sensitive camera on the robot can see where the light spots are, allowing it to easily determine its location relative to the projector, so it can basically create a map of its environment using projector-relative coordinates.

Anybody know of a good way to get perfectly repeatable accuracy to within... say 3-6 inches within a local area? (It doesn't have to have any bearing on actual location, so long as you can repeatably get within a couple inches of the same value at the same location every time you try.)

It wouldn't be cheap, but it looks like the NorthStar system from Evolution Robotics can do pretty accurate robot localization.

These are two companies I've heard of from Idealab where I used to work:

Newbury Network's Location Server products uses 802.11b signal monitoring to do location detection. They offer a virtual docent system that uses the technology to allow the virtual docents to provide location-appropriate information.

Evolution Robotic's VSLAM lets robots use odometry AND visual data to update its position information.

Depending on the size of the warehouse and the manner of "occlusion" that occurs, I'd say wiring up the warehouse with lots of cameras and triangulating to a beacon would be a pretty straightforward method... The hockey puck, indeed!

It has to find and register the wheel, which is an object of known form. Lane Hawk could do this. It then has to find and track the ball, which is not too hard (try the Lucas-Kanade feature tracker in OpenCV) and extract position and velocity. Given that information, prediction is possible.

Now that 3D game capability is going into camera phones, there's enough processing power in phones to consider this. It can all be done with passive sensors. You don't need lasers.

Can do a lot of these tasks and its $299 (supply your own laptop).

Info here

Better yet, for about the price of a Roomba, you can get an ER1 with a camera and robotics-oriented software. An ER1 would make a much better platform for a tablet-PC based robot, since that's basically what it and its software are designed for.

I hope it works better then those laptop-robots.
Robots

The ER1 from Evolution Robotics isn't a bad alternative. You can get one for as little as $199 via their garage sale.

The Aibo Kennel Club has an Aibo AI Mind that needs some re-working and hacking so as to be installed in your pet robot dog -- making it an intelligent companion.

A Python Software Development (SDK) is available for the Evolution Robotics ER1 Personal Robot -- why not for the Aibo?

The Python AI Weblog is a start for coding Python AI minds for Aibos and other robots.

The Visual Basic AI Weblog promotes AI coding not just for Aibo pet dogs but for humanoid robots.

The Perl AI Weblog is for perl geeks, Aibo owners, and coders of server-side 'Net-roaming AI.

The Lisp AI Weblog goes beyond Aibo and other toys into the most serious AI theory and practice.

Please support public-domain open-source artificial intelligence for Aibo and all other robots.

We're currently evaluating what distribution we want to use moving forward.

Well, you could install just about any distribution on a laptop and hook it up to one of those Evolution Robotics laptop robots. Those go forward (and backward, and sideways) quite easily. Oh, you meant in the future? Well, why didn't you say so?

Pointy-Haired Boss: "We need to do this on a going-forward basis!"
Dilbert: "Thanks for ruling out time travel. You're usually not that helpful."

Cheers,
IT

"Robie the Robot" appears to be nothing more than an Evolution ER1 Robotics kit, which Evolution Robotics has been selling for quite a while now. It is a robotics kit that allows you to take an existing laptop and hook it up to some motors and a webcam and control through some command line API's or a nice GUI Evolution has built.

The American Sign Language translation glove was actually introduced at the 2002 Intel Science Talent Search competition by Ryan Patterson of Grand Junction, CO. Patterson's glove uses custom designed electronics to detect hand and finger movements and translate those movements from ASL into their English forms, letters and punctuation.

Now don't get me wrong, I'm not bashing Microsoft or saying that they are ripping off other people's ideas, but if they are trying to bill these items as new research developed at MS R&D labs that's wrong. If they are merely taking these ideas and refining them for future use in the consumer/professional world, then I'm sure that these concepts will benefit from having Microsoft's resources. I'm merely trying to point out that these ideas aren't new in any way, and they have already been conceived and engineered by others, who should recieve all due credit.

...the site listed by /. is slashdotted so here are some useful links. In a nut shell I am not sold by this thing because it lacks a real purpose(for myself anyways). Maybe one day I will move into a 10,000 sq foot house and have plenty of hallways, laptops and time to play around with the ER-1.

http://www.siliconvalley.com/mld/siliconvalley/461 9368.htm

http://www.evolution.com/product/consumer

Now, I just need to know what Battlebot weight class I'm in with my new robomonster!

Seriously though, it's base set-up doesn't look very rugged, nor stable... I doubt your laptop would survive the first fall down a couple stairs.

It's still wicked cool for the price however.

Your laptop can be a beer-fetching robot

REVIEW: Robot Is Fun but Not Useful

By MATTHEW FORDAHL
AP Technology Writer

The ER1 personal robot won't make coffee, pick up the newspaper, vacuum the floors or even walk the dog. With prices starting at $599, it seems most adept at emptying wallets.

But beyond the not-so-cheap shots, the robot has a lot to offer. Like the personal computer kits of the 1970s, much can be learned -- and perhaps someday much money can be made -- on the road to usefulness.

And don't forget the prestige from being the first on the block with a robot smarter than Sony's canine-wannabe AIBO.

The ER1, sold by Evolution Robotics Inc., resembles neither a dog nor the robotic stars of science-fiction movies.

The 2-foot-tall, 20-pound machine is a three-wheeled platform that holds a laptop, its brains, and has a staff that carries a Web camera, its eye. Some might confuse it with an industrial table.

All parts are included except the most expensive -- a laptop running the Windows operating system. Plan on spending at least another $1,000 if you don't already have one.

For the mechanically challenged, Evolution sells an assembled robot for $699 (still minus the computer). But anyone who opts to plunk an extra $100 for the assembled version is missing half the fun.

The ER1 is more of a hobby than a toy. It's not recommended for children under 14, unless they're supervised. Schools might be interested in using the robot to introduce the basics of robotics and programming.

It took about two hours to assemble my ER1, which came in dozens of pieces tightly packed in a box along with 100 screws. Two Universal Serial Bus cables plug into the laptop.

Once installed on the laptop, the software shows a live shot of what the robot's camera sees, various behavioral options and the robot's battery levels.

The instruction manual is especially well done, rare for a high-tech product. It clearly outlined all 32 steps to finish the job and made sense out of the various trusses, gussets, set screws and U clips.

After a few hours of charging the battery, we were ready for our first test -- a routine in which the ER1 recognizes its box and moves toward it.

My ER1 immediately recognized the box but instead of driving toward it, it backed away as though it had been abused at the factory. (Turns out the camera pointed in the wrong direction. The test worked fine after I had adjusted it.)

Such tricks -- including most of the other 50 or so suggested in the manual -- are neat for showing off to neighbors or entertaining at parties. Besides following its box, the ER1 can play music, sing when it hears a loud noise, teach words to a parrot and even warn that it spots a beer can.

The recognition scheme is quite impressive. It could tell the difference between different denominations of currency. It even recognized me as long as I was wearing the shirt I had on when my image was originally captured. It didn't know me from Adam when I put on another shirt.

But the true power of the ER1 is in the ability to layer programs on top of one another, leading to more complex behaviors. Users familiar with the scripting language Python can create even more complicated tasks.

The possibilities are limited only by imagination and hardware. Evolution also plans to sell expansion kits, such as a gripper ($199) for grabbing that beer, and infrared sensors (price to be determined).

Evolution says the 12-volt rechargeable battery that powers the robot's motors can last up to three hours. My laptop battery died long before that.

I had the most fun driving the ER1 around using my home wireless connection. Because my laptop is wireless-capable, I could control the unit from my desktop computer and see everything the robot could see through that computer.

In fact, I could have controlled the thing from anywhere in the world over the Internet, provided I left a few holes open in my firewall.

------

On the Net:

Evolution Robotics: http://www.evolution.com

This definately is interesting, but far too expensive. At $600-700, plus a laptop whose specs must at least meet:
Pentium 3, 500MHz+
Windows 98

I don't think it's a viable alternative to getting yourself a beer.

Why?

• bot's brains are powered by an open source version of this anemic, vertically and aesthetically challenged but smart bot.
• bot aint got none smarts you like? Then twittle some bits. (holy open source batman!)
• bot doesnt look like a dragon (want that? hire your mother in law - more expensive perhaps but meaner too)
• bot embodies all that is Mac. In fact that's what it is mostly.
• bot's drive train is similar to the Segway, its quite mobile and can run down any lesser charged bad person. (But it would'nt - it's a pacifist, you see - it would release the underfed rot-weiler boogie man visceration system. Bot believes in using the right tool for the job.)
• bot knows how to charge itself [bot could learn us a thing or two]. With object recognition it just finds a jack and plugs itself in.
• Bot doubles as a...
  • Mobile desktop/workstation (eg, 'bot, set up my workstation out there yonder by the olympic pool next to Heather - dont forget the beer this time. Oh and blow Heather up too')
  • Mobile access point
  • Mobile entertainment system
  • Door man.(If the guest is carrying a bible or a gun - other appropriate action is taken ;-))
  • Art
  • A writer. Bot write good, master.

Have a look at Evolution Robotics. They sell hardware that gives wheels to your laptop. It also comes with software, unfortunately Windows-based (they also have a Linux version but they don't ship that).

For turning, extend the dead reckoning code to also keep track of the rate of turning. Then you can also handle the situation where a crossing shows up in the middle of a turn, by continuing to turn at the same rate while crossing.

Next, it needs a web interface, an OGG player, cam and a robotic arm to turn on the coffee machine, and... oh wait, that would be the ER1 from Evolution Robotics. I'm not sure if it can turn on the coffee machine, but it can deliver your pizza and hunt down people!

Well, if the wheeled base is connected via USB and has a control pannel, then a USB arm seems just as possible. Not that I could do it, though.

Put arms on it! Send it to pick up the paper and pick up dog crap. That's what robots are for.

The RDK (Robot Development Kit) page says you can get a laptop preconfigured to run this thing. The laptop runs "Standard Red Hat 7.2". Cool!

The thing has loads of sensors: Video input, Sound input, 9 infrared sensor and 4 bump sensors.

Wonder if the hardware interface has an open spec? Again, not that I could do anything with it. Not for a while.

This is incredibly cool.