Slashdot Mirror

I figured a Battle Roomba would look more like Ziggo.

In response to this recent product announcement several middle eastern countries have purchased large quantities of the Roomba Virtual Wall accessories and it is rumored that these are being installed along territory borders.

The company I-Robot uses Java to interface a Linux box running Red Hat to all the sensors, cameras, and motors on their research line of robots. I spent four months trying to get the darn thing to work, CLASSPATHS and all to no avail. What a friggin nightmare. My master's thesis was going to be on that robot and now Auburn University is just building a robot from the ground up. Life's like that ...

iRobot is a cool company; if you need replacement parts for your Roomba, the friendly people on the phone are more than happy to mail them out post-haste to you.

So the new Roomba models head back to their base to recharge themselves. That's really cool. But can they be put on a timed or daily schedule? That would address the top two questions I got asked about my first-gen Roomba.

~jeff

Robotic servants... you mean like Roomba http://www.irobot.com/consumer/? Or those hospital delivery robots http://slashdot.org/articles/04/07/06/217238.shtml ?tid=137&tid=216?

Does Common Lisp lend itself easily to kernel, systems, or real-time programming on mass-market computer architectures?

I know that the developers of a popular PS2 game have disclosed that the game's business logic is written in Lisp

but would one want to write a 3D engine or a cellphone game in Lisp?

PS - in a recent dick-measuring benchmark (the "Coyote Gulch" floating-point ephemeris calculator), CMU Common Lisp and Steel Bank Common Lisp produced code 5% faster than GCC on a Pentium IV. The participants claim it was "an hour-or-so's work".

And what stake does the Roomba company have in this?

I am in robotics and I know of a company called irobot. I just believed it was their new product. It was a surprise for me because I do keep track of advances in robotics, and had never heard of this product. But the company irobot has smart people, so I was willing to believe it.

How exactly do they expect people who have never read anything by Asimov to catch on that this is a movie?

The reason the movie website is 'irobotnow.com' is that iRobot is the folks that make the Roomba vaccuum, and military robots.

So well, if Apple had an issue, they'd have already been bitching.

The machine I think should be better described as a wall painting robot rather than a "wall printer".
A similar thing would be to fit a spray can onto this wall climbing robot.
Surprising that nobody has automated wall painting yet, then we wouldnt need to have these plain colored walls anymore. All of us could have our favorite frescos. Just get me a large res. image of the cistine chapel roof!!

Historically, AI has cycles of fads. Somebody has a reasonably good idea, they claim strong AI is right around the corner, there's a few years of frantic activity, the new approach hits a ceiling, and the fad is over. Major fads have been backtracking, LISP, perceptrons, theorem proving, expert systems, neural nets, genetic algorithms, and reactive robots. Each of those hit a ceiling after the first few years.

Much of what's going on today involves building systems that give the impression of being more intelligent than they are. Ask Jeeves is a good example. It has no idea what your question means, but finds results that have some vague releveance to what you asked. Lenat's Cyc is like that. Lenat's 1984 paper, "Why AM and EURISKO Appear to Work", shows Lenat in an honest moment, admitting that what came out was mostly implicit in what had been built in. Cyc itself has been "close to success" for a decade now. Most recently, Lenat got some "homeland security" money to use Cyc for data mining, another application where understanding content isn't necessary.

In a different direction, there are systems built inside dolls to give the illusion of humanness. Brooks has built a few systems like that, including My Real Baby, an unsuccessful Hasbro product. Brooks did good insect-level robots, then started claiming that reactive systems were going to lead to strong AI real soon now, and started Cog, which is a humanoid torso that reacts to people nearby but doesn't do much in the way of useful tasks. I once asked him why he didn't try for mouse-level AI, as a next step after insect-level AI. He said "because I don't want to go down in history as the man who created the world's greatest robot mouse".

That's a big part of the problem - hubris. Trying to get to human-level AI when we can't do lizard-level AI isn't working. Yet spending your whole life trying to build a good lizard brain isn't a good career move.

The one bright spot is in gaming. Game developers need lizard-level AI (balance, running, survival, fighting, etc.) and they're slowly making it work. If you want to see real progress in AI, go to GDC, not AAAI.

AI needs to focus more on tasks where failure is possible. This keeps people from faking it. We're currently putting together a vehicle for the DARPA Grand Challenge. 200 miles across the desert, no driver, no human intervention, no question about whether you succeed or fail.

I asked Rod Brooks if the MIT AI Lab was entering. He said no.

iRobot's taken....

So get a Roomba.
It does not look like a person, but who cares if it still does your floors.

No suggestions for the beers yet.

There are movies on the iRobot website, some of them are interesting and definitely give you better perspective of exactly what kind of results are acheived through the usage of the swarming idea. The movies are available here.

There are movies on the iRobot website, some of them are interesting and definitely give you better perspective of exactly what kind of results are acheived through the usage of the swarming idea. The movies are available here.

Holy brushes with techno-greatness, Batman! I was at iRobot last week and met this guy. He was driving around "a swarm" (actually, the lead of the swarm if you wanna get technical) of robots that were cubes about 4" to the edge.

They were VERY similar to what you see here but cube-shaped.

The thing I remember most about the guy was that he was wearing gigantic yellow killer bee slippers. Talk about a laid back place to work. Very cool.

A friend has told me of a cat that likes to ride on one, but I haven't seen pictures yet.

My question is if, as the article says, the thing can use ir to follow walls, why does it get its nose stuck under the cabinets in my kitchen that are just the wrong height? (it will eventually usually unstick itself, but it could be a little smarter I think).

Still, it works way better than I thought it would.

For those robot geeks among us who did NOT know, this is Rodney Brooks' company.

Rodney A. Brooks is Director of the 230 person MIT Artificial Intelligence Laboratory, and is the Fujitsu Professor of Computer Science. He is also Chairman and Chief Technical Officer of iRobot Corp (Roomba)

He received degrees in pure mathematics from the Flinders University of South Australia and the Ph.D. in Computer Science from Stanford University in 1981.

This guy is to robot-geeks what RMS is to Open-Source.

iRobot.

I expect that robots will take over the world, and openly hunt humans in a post apocolyptic landscape. This will occur in January.

Funded by DARPA = Eventual military use for this...

So what exactly is this for, remotely wardriving in Afganistan?

There is a program at known as Future Combat Systems. One of their big things right now is teleoperation technologies. They are looking at a whole school of Unmanned X Vehichles, where X is both arial and ground vehicles. At my work, we've been collaborating with the Mobile Robotics Lab at Georgia Tech, turning a Hummer and some robots known as ATRVs into teleoperated bots. We have been doing this as part of the communications portion of the Future Combat Systems project, to demonstrate an IP based communications network developed by another company. Last week we drove the ATRV from New Jersey while the robot was in Atlanta. The hummer can now be driven over telnet, and probably can be driven over a similar distance (although safty concerns make testing such things a little more difficult, and caused us not to try). We can drive any of the robots by gaming joystick from a computer on the internet, with video latency being the limiting factor. And yes, all ye Linux zelots, all the computers in the project run Linux, except an old PC-104 stack running Dos from a floppy.

One think that I have picked up is that just because DARPA is currently looking at things, it does NOT mean that they are making any of them. DARPA will from time to time fund things like this just to find out what the "Best Effort" of industry is, that way they know exactly what they CAN have made.


To get my email address, add "@mail.gatech.edu" to my slashdot ID.

As a member of the team working on the Lewis project, I'd like to provide some additional technical details. It should also be noted that the Lewis project is not intended to replace human photographers. It's an easily accessible research-oriented endeavor to explore human/robot interactions in a real-world environment.

Specifications:

• Pentium III 800Mhz CPU
• Linux operating system, kernel v2.2
• Wireless ethernet
• Sony DFW-VL500 digital 1394 camera
• Approximately 4' 6" tall, 2' diameter, 300 lbs with batteries
• 4 12V deep cycle lead-acid batteries provide nearly 6 hours of continuous use between charges.

Processor - Lewis is a B21r mobile research robot from iRobot Corporation. It's powered by a single 800Mhz Pentium IV processor. This CPU must handle all of the motor drive and low-level robot tasks such processing the data from the large array of sensors. On top of this CPU load is the task of finding faces, navigating crowds, and taking and processing the photos. The two additional processors to be installed in the future will allow Lewis much more power for its photographer duties.

Camera - Lewis currently uses a Sony DFW-VL500 [technical manual] digital 1394 (Firewire) camera. This has a 1/3" CCD that produces 640x480 color images at up to 30 frames per second. Image output is YUV 422 format and is not compressed. The built-in 12X zoom lens is sensitive to 14 lx (F1.8). Higher-resolution 1394 cameras are available, but these do not have built-in lenses; this is bad because focus, aperture, and zoom must be fixed.

Safety - The entire enclosure is lined with bump-sensitive panels, so that if the robot runs into anything, the currently executing program is terminated, the motors are halted, and the brake is applied.

Operating System - The operating system on Lewis is a standard Linux distribution using kernel version 2.2. Various libraries for control of the motors, sensors, pan/tilt unit, and camera are used.

A couple of other comments: the camera is not an NTSC video camera. It was chosen because of the easy ability to control zoom, focus, and aperture from software. Since our goal at the moment is not film-quality pictures, this camera suffices.

Sample photos are available on our website. We have been slow in posting samples due to privacy concerns, not because the pictures are bad. We have over 3,500 photos, and I'd say well less than 2% are false hits -- photos of doors, walls, elbows, etc.

This whole wedding photo thing is overrated. The last wedding i went to wasn't bad, but the first time one of my friends got married after college, there was like an hour of picture takin' downtime between the ceremony and the food. This whole process could be done away with by a high-end image based rendering package. You could get a few good images of each guest as they came in, maybe a 3D laser scan, and a few really detailed datasets for the bride and groom while they're being fitted for the dress and tux (or whathaveyou). Render up a few group shots with the romantic background of your choosing, some closeups of the couple, and you're set. If you wanted to, you could add capabilities for candid reception shots (although you might need some training data for the system to transform 'respectable wedding guest' photos into 'drunken reception partygoer' photos). Plus, you don't have to worry about real life situations like that one 8 year old who refuses to smile for the camera, and you can do Orwellian modifications if you later decide you wish you hadn't invited some particular person.
Or you could just hire a decent photographer, maybe do the time consuming stuff ahead of time, and leave a bunch of disposable cameras around for the reception.
Not that i can in any way fault the robot idea. It does, in fact, kick ass.
(And one last nitpick, it's not really fire engine red, it's Research Robot Red. If it's anything similar to what ActivMedia uses on the Pioneer robots, it's an amazingly impervious, apparently epoxy-based coating which can only be removed with a grinding wheel or similarly serious abrasive.)

Their best known product is My Real Baby, manufactured by Hasbro around 1999-2000. It's basically a baby doll with Furby-type software. Rated "Worst Idea of the Year" by the Alliance for Childhood. It's not even that original; Baby Think it Over, the anti-teen-pregnancy doll from hell ("requires real care on the part of the student, including feeding, burping, rocking, and changing diapers"), has been around for years, but at a price well above the toy level.

This whole direction is way too much like Eliza. Much of the AI field, having failed at tasks that actually require doing something successfully without human assistance, now seems to be focused more on faking it. You've all seen Ask Jeeves, and obnoxious "virtual customer support reps". Those are pathetic.

There's some good work going on, but this isn't it.

May be closer than he thinks.

The same company, iRobot, makes these
Dolls.

You might check out iRobot. I work at the MIT AI Lab and we use the B21r for my project, which has similar goals. This robot was originally designed by a company called Real World Interface (RWI) that was ultimately purchased by iRobot. In any case, the robot runs Linux and uses V4L as far as I can tell, with what appears to be an off-the-shelf Hauppage WinTV card or equivalent. I don't know much about the camera, but conceptually, you should be able to set something up very similarly.

Now, as far as actually writing the code for real-time vision... that's up to you - good luck - it is a difficult problem.

Those things are not easy to drive. One of the most difficult things is getting a perspective on where the robot is in relation to it's surroundings (very rough rubble). This is an ongoing research area for many robotics teams, and one we have been working on also.

With a simple black-and-white camera it is hard to tell what's out there. Is there something behind that grass? Is it a rock, a hole? or a log that can high-center us? Now roll in trying to do that autonomously. The computer can see the clump of grass, but if its thick enough, it'll see it as an obstacle instead of something it can just roll over. And even if it knows it can roll over it, it can't see if there are any obstacles hidden in the grass. We're trying to use laser scanners and radars to try to solve that problem.

Multiple cameras helps, but adds significant complexity and disjoint views. A technology which really makes this easier is an Omni-directional Video sensor (which has a 360 deg. field of view around the sensor).

Yup, we have one of those from a company called Remote Reality. Nifty items. The one with the hole in the center of the picture gives the best resolution at the horizon. 180 degree fisheyes don't give the hole in the center, but suck at resolution on the horizon.

--Carlos V.

Anyway, the field of robotic USR at this point is so new that a lot of interesting and very different technologies are being developed. A lot of groups use platforms that wouldn't be of great use in a real rescue situation, such as the iRobot Magellan Pros we used, simply because they're research robots designed to roll around on flat surfaces. These in particular lack sensors specifically designed for rescue use - they have a ring of sonar and infrared range sensors around them and a video camera. Others, like the Urbans used at USF, are very rugged (they are tracked and can climb things) and have very sensitive FLIR sensors for detecting people.

Autonomy versus teleoperation was the subject of some debate at the competition last summer. The scoring for the contest was done according to an equation that rewarded for the number of people found, penalized for the number of people it took to operate the robots, and so on. The formula gave a very clear advantage to teleoperated robots: even though teleoperating is very difficult (it's not like RC cars!), it's a lot easier for a human to look at a screen and navigate/find people than it is for a robot to derive this information from the same image. In a way, this is a very pragmatic approach: we need to develop useful technologies soon. On the other hand, this contest took place at an AI conference, where naturally advancing the state of the art of machine intelligence is viewed as a pretty important goal.

Because of the robots' differing abilities, the contest featured three seperate NIST-developed standard courses; one with flat floors, one with more challenging debris, and one with what essentially was a big pile of rubble. Our Magellans could only handle the first room; the Urbans and a huge homebrew tracked robot from the Sharif University of Technology in Tehran took the more challenging rooms.

In any case, here (at last) is a summary of some of the technologies used in the competition:

• Swarthmore College (us): Standard Magellan Pro wheeled research bot with a Canon pan-tilt-zoom camera on top. Final version used "semi-autonomous" guidance, where an operator would tell the robot to go to a point not too far away and the robot would go there. Big lead-acid gel batteries; PC on board did lots of processing. <gimmick>Robot could also generate red-blue stereogram images of site on operator command by rotating slowly and compiling distance data.</gimmick>
• University of South Florida: Teleoperated Urban tracked robots with FLIR sensors. Some info. Probably lead-acid batteries, don't know about onboard processing (suspect no much).
• Sharif University of Technology: Medium dog-sized tracked robot with binocular vision. Teleoperated by laptop or Palm. I wouldn't want to get in the way of this thing. Few or no sonars IIRC, giant lead-acid gel batteries, probably very little onboard processing.
• University of Minnesota: Cute tube-shaped robots designed (no joke) on a DARPA grant to be shot out of a grenade launcher. Completely radio controlled; AFAIK no smarts on board. Sent wireless video feed to monitors. A spring mechanism allowed them to jump out of tight spots. DARPA doesn't let them talk about power supply, other info.
• ?University of Utah?: A commodity approach: robots were Radio Shack RC cars with Basic STAMPs on board. While not completed at the contest, eventually they will swarm with complete autonomy around the disaster site, detect people (or other hot things) with IR sensors, and relay their position to a central controller. Power is a simple 9V battery.
• University of Edinbugh: don't recall exactly; despite all their valiant efforts, they couldn't get anything working in time. I think they were also working on an autonomous system of two or three bots.

Enough rambling for now. In any case, it was a really cool experience to go out to Seattle and see all the stuff the teams were trying. For me, robotics is a great, growing field that is a whole lot of fun, and the conference was one unforgettable week.

--Tom

It may be "the world's first attempt to sell a humanoid robot that users can program freely", but it surely is not the first robot that users can program freely. Researchers at universities want robots that have all/most of the hardware or software specs open. Like data sheets of electronic parts and source code of control software. And this has been the case for a long time. For example, these research robots have always been freely programmable.

In this "robot".

In /. for posting this story.

Why?

This has to be the lamest robot "story" on the planet. Similar "security" robots have been built by companies and individuals for years. I remember several companies in the 80's doing this, when robots were the "thing" of the "future".

GPS? Why GPS? A white or black line (or even one done in a flourescent "invisible" paint or something) would be much cheaper for navigation. In a new building, a buried wire under the floor or carpet could be used. Coded tags at doors could further aid navigation (UPC or IR "active" tags).

Nomadic Technologies used to sell research robots with this kind of use in mind (sadly, I just found that they stopped production).

IRobot has a research robot that seems ideal for this as well.

Of course, nothing would beat Odetic's Odex-1 for the "scare" factor in security - too bad this 80's robot never went into production...

Now, homebrew bots - that is where the action is:

Karl Williams seems to have many projects of the type that would make interesting security platforms - or at least something to build off of (mount the vortex cannon or coil gun onto the home drone - yikes!).

This machine might even be better for security - simply because it could be smaller and faster for such a job.

The truth is that there are a lot of homebrew and commercial robots that can easily do what this robot does - probably at a fraction of the cost (actually, some of the commercial bots are quite expensive). There were many robots built in the 80's that were capable as well.

That is the article I want to see. Somehow I was hoping for a two legged chicken walker (not ED-209 sized, but something) patrolling the halls, maybe packing low powered pea shooters for "defense" (actually, one homebrewer managed to build such a robot with a "pea-shooter"-style, multi barrel "gun" - it couldn't hurt you, but it could knock over empty pop cans - I wish I had a link to it - probably do, but it is buried in my link list somewhere deep).

Oh well...

Worldcom - Generation Duh!

Rodney Brooks' crew at www.irobot.com have got a remote "avatar" robot equipped with web cam for exactly this purpose. Highly neat!

They didn't mention the iRobot. This thing was shown at Comdex. Its robot that is controlled via a web browser. The robot runs on linux, uses a wireless connection.. there are a bunch of specs at the site.

CNN has some coverage of the irobot here.

-gerbik

I just got done reading an article in this months' issue of Popular Science that discusses three or four up-and-coming robots that will be commercially available. They're a bit pricey at around $4K a pop, BUT...one in particular (irobot, www.irobot.com), even runs LINUX as its OS.

A great gift, the iRobot, it runs off of Linux 2.2, with a K6-2 450 Mhz CPU, 6GB IDE harddrive,runs Apache,has a wireless internet connection, built in webcam,can climb stairs in 10 secs, allows for remote access through the internet, has a SDK coming out soon, an over all great geek gift, all for the low, low price of $4,995, but still its a great gift.

When I saw the headline, I was hoping this article was about the amazing iRobot. That's what I really want for Christmas!