Imagine you’re a hunter-killer robot, hovering over the broken wasteland that used to be the world of men. You have surprised a group of biologicals in an act of petty insurrection, and they have split into two groups and begun to flee. You can only pursue and eliminate one group, but you don’t have the spare milliseconds to analyze your high-definition imagery; yet you must determine which group is greater if you want to meet your termination quota.

It’s a good thing that back in 2012, a university lab in Italy helped machines like you evolve approximate number sense!

Yes, Marco Zorzi and Ivilin Stoianov have opened the Pandora’s box of guesstimation in their shadowy “laboratorio” at the University of Padua. Or rather, it opened itself, as the ability seems to have emerged naturally from basic learning processes and not from any programmed understanding of numerosity.

Seriously, though, this is very interesting research. Zorzi and Stoianov created a virtual neural network simulating a basic retina-like structure that “sees” pixels, and then two deeper layers of nodes that sort and analyze the input from the “retina” layer. Strictly speaking, the retina is already composed of several layers with various levels of analysis, but we’ll let that go for now, for science.

The self-revising neural network model they used (in other words, a small-scale, learning AI) was not given any lessons on numbers — it did not know the difference between 2 and 4, integers or fractional numbers, higher or lower numbers, anything like that. This was intended to mimic the early stages of development in creatures that demonstrate ANS: approximate number sense, described as the ability to determine basic numeric qualities such as greater or lesser without actually understanding the numbers themselves. Infants demonstrate it before learning basic arithmetic, and fish demonstrate it when choosing larger and therefore safer shoals to swim with, when presumably they are not counting their colleagues’ numbers exactly. Many other animals show it as well, in situations you can probably imagine.

Zorzi exposed the neural network to a series of 51,800 images comprising various numbers and sizes of rectangles spread around a field. The network attempted to recreate in its own way the images and determine rules governing them. After a number of exposures, the network exhibited on the “lowest” level of neurons (i.e. the most meta-analytical) that there were some neurons firing more or less in correlation with the number of rectangles, but not the total surface area taken up by those rectangles, indicating that the AI was detecting numbers and not simply contrast or the like. Remember, this AI doesn’t even know what numbers are.

They further solidified their findings by letting the computer estimate whether a given image had more or fewer objects than a given number. It had indeed developed a rudimentary ANS system. Interestingly, there are actual neurons in the parietal cortex that exhibit this same behavior.

This type of research typifies the next phase in the real-world/computer interface: natural learning and fuzzy logic. The ability to take what has been detected and turn that into a system of rules, much like the way our own minds are formed, is going to be increasingly important. There might be little things like — your Kinect hears the front door open between this and that hour, and automatically turns on the TV and queues up the show you always watch on that day of the week. Or a security camera learns the faces it needs to pay attention to, or a helper robot learns when to follow master and when not to based on cues the creators might not have foreseen, or anything else you can think of.

Basing our devices on ourselves is one of the ways to make them easy to relate to, though it’s not a guarantee that they will be useful. There’s no use training up a car-painting robot from babyhood, but it would make sense for more domestic devices. Having our devices think and act like us is a natural path, though, and while at the moment our resources seem to limit us to simulating functions we as humans (and fish) develop in the cradle, that doesn’t mean more sophisticated abilities can’t or won’t be developed. Indeed, it is easy to underestimate the sophistication of the most basic functions we as active, conscious beings exhibit every day without even noticing.

The paper, “Emergence of a ‘visual number sense’ in hierarchical generative models,” was published earlier this month in Nature.

The Robocalypse is near: Osaka-based Hajime Research Institute is working on a humanoid that will stand 13 feet (4m) tall, which is much taller than most other existing robots of its kind (the latest version of Honda’s Asimo, for example, is just 130cm high). It will also sport a built-in cockpit.

Hajime Sakamoto, president of the institute, is on a mission. After having successfully developed a 7-foot robot in 2009 (pictured above), the plan is to build the aforementioned giant robot next, before following up with humanoids that are 26 and 59 feet tall after that.

Needless to say, the 13-foot version would be the tallest humanoid out there. Sakamoto is currently trying to find more contractors and sponsors to finish the project, for example by posting videos on YouTube, setting up a Facebook page, or running a blog (all in Japanese).

Sakamoto doesn’t say how long it will take to build the robot, but this picture from Monday shows him standing him next to a completed leg:

Via Plastic Pals

And we’re one step closer to the Robocalypse: a research team at the Nagoya Institute of Technology has developed a passive robot that can walk on its own. All it needs to get moving is a gentle push – no motors, sensors or even electricity needed. Its legs and ankles are powered by potential energy only.

The makers says the robot is made of aluminum and took 100,000 steps in a row during a test (which lasted 13 hours). The robot is able to carry a payload of “10-20kg” and could be used by people who have trouble walking, for example (it’s possible for humans to “wear” the robot on their legs).

The robot is 90cm high and walks at 3.3km/h. A commercial version is expected to become available in 1-2 years.

This video from Diginfo TV (shot in Tokyo, in English) shows the robot in action:

I’m sure all of you remember BigDog, the utterly terrifying biomimetic robot platform that dances over rough terrain on weird legs and makes a sound like it’s full of angry bees. You didn’t think they were going to give up on it, did you? No, it’s been a rousing success apparently, so much so that they decided to build one that outdoes the original in every way.

The AlphaDog, as they’re calling it, was created to be basically the practical version of BigDog, which, while compelling and scary, was too loud and couldn’t carry enough stuff. AlphaDog, on the other hand, will carry 400 pounds of gear for 20 miles in a day, without needing to be refueled.

Be sure to watch the video to the end. The now-traditional “shoving of the quadrupedal robot platform” is great, and then you get to see a new creepy behavior as it rights itself in an unreal way from a lying position.

Right now it’s powered externally and running on rails, but it’s not so hard to think of this thing in the field. With a GPS unit and some rudimentary depth-sensing cameras, it could confidently navigate most terrain or follow its “masters” along a trail. Though I doubt that feature will make it into first production AlphaDogs, scheduled for 2012.

Boston Dynamics’ webpage doesn’t have too much extra info, so keep an eye on their YouTube account for news.

The huge variety of robots is a thing to marvel at. Just a few years ago, the best they could do was roll around or walk stiffly on poorly-hinged limbs. Now we have robots that gallop, slither, and fly like a bird. This mode of transportation is totally new to me, though. The Paraswift climbs up walls by generating a suction, then gets itself safely down by jumping off and deploying a little parachute. Why not, right?

Check out the video. The action happens about halfway in.

Paraswift is a collaboration between Disney Research and the Swiss Federal Research Institute. This demonstration took place during the International Conference on Climbing and Walking Robots, held at the University Pierre and Marie Curie in Paris.

It generates the suction by creating a vortex “like a mini tornado” in a central tube; the vortex has a low-pressure center, and this pressure gradient, through a process known as sorcery, causes the robot to stick to the wall. Only the wheels need to be in contact with the climbing surface, which makes it much more useful for climbing surfaces like stone and concrete, which can be difficult to form a vacuum seal around.

Once it’s had enough climbing, the little wings deploy, spreading the glider/parachute, and it jumps to relative safety. I say relative because it looks like there’s a bit of a bump at the end there. They’ll work that out in version 2.

Unfortunately, the first use I can think of for these things is setting them to climb up apartment buildings and assassinate the fleshy humans within. So this one gets filed under Robocalypse.

Japan-based tech startup ITK has brought us one step closer to the Robocalypse. Roboticists around the world are working on manufacturing “sensitive” hands for robots suitable for touching humans or handling breakable objects, a problem that’s notoriously difficult to solve.

ITK is now throwing their hat into the ring with Handroid a new model that seems to be one of the most advanced robot hands out there. Sporting five fully movable fingers, the Handroid looks a lot like the hand Arnold Schwarzenegger uses in Terminator 2.

Operators can control each finger remotely (the Handroid can mimic the operator’s movements), for example in environments where it’s too dangerous to use human hands. At 740g, the Handroid is very light, prompting ITK to look into possibilities to develop a prosthetic hand that could be controlled by brain waves or cerebral nerves.

ITK plans to commercialize the Handroid in about two years for about $6,500 per unit – a very low price for such a device.

Here’s a promo video:

Via Robonable [JP]

It’s been a while since we’ve posted any interesting Kinect hacks. They were coming fast and furious for a while, but this is the first I’ve seen in a while that really makes me happy. Robotics students at Stanford put this “JediBot” together from an articulating arm and a Kinect (and of course a little robotics know-how). The Kinect tracks the location of both “lightsabers” and has the robot make an attack or defending move based on their positions.

Right now it doesn’t go too fast, but considering what these robot arms are capable of, I’m not sure I’d want it going at full speed.

Check out Stanford’s news item on the project here, then move on to the Robocalypse tag to see why teaching robots to sword-fight is not the best idea.

Worried/excited about the increasingly powerful and versatile robot army we’re creating? You’re not alone. Spend some time perusing the Robocalypse tag and you’ll be more interested/terrified than ever.

[via Robots.net]

For now I guess we should congratulate Nikolaos Papanikolopoulos and his team at the University of Minnesota for creating a pretty cool little bot. Their design is solid, although right now there are two models, one that only flies and one that only rolls, since they haven’t worked out how to get the motor to power both drives.

The next step is to make it simpler and lighter, and making the robot navigate autonomously instead of by remote. Isn’t it a little early for that, guys? Don’t need to bring down humankind any faster than necessary.

[via Ieee Spectrum]

The Robocalypse, it’s coming nearer and nearer. Case in point: POLYRO (short for “oPen sOurce friendLY RObot”), a mini humanoid that apparently costs less than $2,000 to produce, all parts included. Developed by American robot researcher Timothy Payne, the little guy is meant to be used to explore human-robot interaction and can be built following these instructions.

The humanoid, in its current form, is made of a total of eleven servos, two web cams (one for each eye), a 10.1-inch netbook, a Microsoft Kinect, and a number of other parts (scroll down on this page for a full list).

POLYRO stands 99cm tall, weighs 8.6kg and runs on Linux.

Via Plastic Pals

Unfortunately I can’t embed the video here, but you can watch it at ABC in glorious 400×220 resolution. It really is a very interesting little segment. And hey, it’s Friday. You deserve it.

This new one? Looks just like a hummingbird, has an onboard camera, and flies for ten minutes.

Watch the video. The device, designed by Aerovironment, has a wingspan of 16cm, or 6.5 inches, and weighs only 19 grams — less than an ounce. Yet it still has its own battery and motor, and can send a live video signal to a palmtop controller/monitoring station.

Welcome to the future, people. And, of course, welcome to the Robocalypse.

[via Hack A Day]

Background from the makers:

The presented approach allows a free-standing, self-balancing humanoid robot to acquire new motor skills by kinesthetic teaching. The method controls simultaneously the upper and lower body of the robot with different control strategies. Imitation learning is used for training the upper body via kinesthetic teaching, while at the same time ankle/hip reaction motion patterns are used for keeping the balance of the robot. During demonstration, a force/torque sensor is used to record the exerted forces, and during reproduction, a hybrid position/force controller is used to reproduce the learned trajectories in terms of positions and forces to the end effector.

And it’s getting better. The researchers say that their “upper-body kinesthetic teaching” method can pave the way to developing robots that are able to handle various “vertical surface related tasks”, i.e. cleaning windows or painting walls.

Here’s the video:

More pictures of Fujitsu HOAP-2 can be found here.

Thanks for the tip to “Tipper Tipper”!

The spokesman for Mayekawa, the Japanese company that designed the robot, noted that the use of robots in food preparation and service lags severely behind the use in other industries. I’d say that’s because we just don’t trust the things to fillet our chicken yet, but there have also been technical concerns. The HAMDAS-R is able to “feel” where it needs to stop cutting and won’t accidentally chop through bones or leave fragments in the meat. That’s the theory, anyway.

I’m sure these guys have been tested a lot and will be backed up by quality assurance people sorting through the meat and looking for obvious problems. Look at that assembly line, though. With teamwork, anything is possible! Actually, that’s what I’m afraid of.

[via BotJunkie and Gizmodo]

That’s Northrop Grumman’s approach, anyway. At the Association of the U.S. Army con in Washington, they’ve showing off a CaMEL packbot with a .50 caliber machine gun controlled remotely by a touchscreen. Of course, like a booth babe, it’s a trick. Wait, that came out wrong. It’s a ruse.

NG doesn’t actually make these things: “Quite frankly, a weapon on a robot brings people into the booth,” says Jon Anderson, who ran the booth. But will they ever? Apparently, the army just hasn’t bought into it yet. Chances are it’s a bit difficult to fit one of these things into the order of battle. And they look pretty heavy; if they break down, you’ll need a non-weaponized one just to carry the first one out.

It’s only a matter of time, though. That’s it for us humans.

One of the things that needs looking into is how fast robots should move. You’ve seen the videos of high-speed robo-arms playing baseball or making ramen — you wouldn’t want to work around those guys; movement a few inches too far could seriously injure someone. So researchers are looking at how hard a robot can hit you before it starts to hurt.

The robot (repurposed from Epson) whacks volunteers on the arm with various objects, with various speeds and forces. The volunteer rates the pain caused — but don’t worry, they’re only testing up to “moderate”; these poor college kids aren’t being savagely beaten.

It’s all going on at the University of Ljubljana. But I can hear the screams from here.

[via SlashGear]

After being instructed how to hold the bow and release the arrow, the robot learns by itself to aim and shoot arrows at the target. It learns to hit the center of the target in only 8 trials.

Does that sound like good news to anyone?

Of course you might say “Devin, we already have sniper-rifle equipped murdercopters. Why do we need to fear a robot with a bow and arrow?” To which I would respond, you traitor to the human race, that you need to watch Predator and First Blood to see just how effective a bow-and-knife wielding brush-stalking hunterbot could be.

As a final note, why do we insist on putting the creepiest faces on robots? Why not something more, I don’t know, robotic?

This robot, among others, will be shown next week at the International Home Care & Rehabilitation Exhibition in Tokyo. Right, rehabilitation.

The MQ-8 Fire Scout, a model being considered for combat deployment, was undergoing testing with a ground crew when its control connection was somehow lost. Instead of circling aimlessly, as UAVs tend to do when they lose signal, it decided to execute part of its program and start flying northwest, entering restricted airspace along the way.

The ground crew managed to catch up and re-establish contact, and of course it’s not like they’d programmed the thing to fire on any moving target — but the event is still a little scary. What happens when a bug like this one (which they just fixed) get deployed to a whole arsenal? I will tell you: Robocalypse, baby.

It actually is reminiscent of an episode of Ghost in the Shell: Stand Alone Complex, one of the best actually (“Natural Enemy”), in which some military helicopters go haywire and the team has to draw them away from a delicate situation. Hopefully we won’t have to snipe the cyberbrains out of anyone just yet, though.

[via The Register]