We’ve written about holonomic wheel systems in the past, extolling their many virtues, specifically their ability to move in any direction without the need for z-axis rotation. Holonomic wheels are just like normal wheels, except they have a bunch of rollers attached to the rim at an axis of rotation at a 45° angle to the plane of the wheel in a plane parallel to the axis of rotation of the wheel. Make sense? No? Okay, video:

Isn’t that cool? This is the RMP 50 Omni robotics platform, developed by Segway. Yeah, that Segway. It allows for rotation and translation in any direction irrespective of the orientation of the platform. The benefits for indoor navigation are obvious, as are the benefits for (if I may say so) the ComBot arena, where maneuverability is key. The base weighs 150 pounds by itself and can carry a 150 pound payload up to 3 miles at a dignified 2 mph. It has a 9mm ground clearance and no suspension, so you’re not going to be able to do much in the way of off-roading. Not that you’d want to… At $21,000, I’d probably buy some sort of lesser robotics platform just to move the RMP around safely.

[ Segway RMP ] VIA [ Hizook ]

Exploration robots generally have two basic functions to perform: taking pictures, and taking samples. Taking samples is the really tricky part, since it requires the robot to navigate unfamiliar terrain, and often the most interesting samples are in equally interesting geological locations, like on the sides of cliffs and inside craters. NASA’s Opportunity rover spent more than a month stuck in a sand dune a few years ago, and Spirit has had similar problems.

One solution to the problem is to design a sub-rover specifically for sample taking, allowing the primary rover to stay safe. The sub-rover would be small, simple, and robust, and NASA engineers have designed a prototype that consists of just one single axle with a sample arm sticking out of it:

It’s hard to see in the video, but there’s a cable running from the center of the robot to an anchor point up the wall behind it. On Mars, the cable would be anchored to a larger robot, or even to a human. It’s only got 3 motors, but using the cable for leverage the robot can climb loose or slippery surfaces, vertical surfaces, or even traverse overhangs. The axle robot is just a proof of concept at this stage, but it’s such an intuitive design that I’m confident we’ll be seeing more of it.

VIA [ YouTube ]

If there’s one thing that’s hard to find here on Earth, it’s the moon. I guess it’s pretty tough to find a good analog for the lunar surface, ’cause NASA has decided to test out a potential lunar rover called Scarab in Hawaii, of all places. Yeah, the volcanic landscape there is kinda like the lunar surface, but the test site 9,000 feet up Mauna Kea also has snow, fog, wind, rain, and 40 degree daytime temperatures, most of which are decidedly un-moony. But you take what you can get, I suppose.

Scarab has been designed by Carnegie Mellon to drill into the lunar surface in search of useful stuff like hydrogen, oxygen, and a convenient mixture of the two called water. It’s able to navigate autonomously around the dark side of the moon, without relying on constant contact with Earth or constant power from the sun. The radioactive isotope power generator on Scarab is good for ten years (that’s ten years), the trade-off being that you can’t get the power out of the generator very fast… It’s like the energizer bunny on downers. The generator outputs 100 watts of power (i.e. a weak lightbulb’s worth), which has to keep the rover’s systems going while it’s either moving around, or drilling. Consequently, the rover just does stuff very, very slowly. But that’s cool, there’s no rush. It’s got ten years, remember?

Video after the jump. (Read more…)

I wrote an article over on OhGizmo today about a vocal mouse interface that the University of Washington is developing for people who are unable to use a traditional computer mouse. The Vocal Joystick uses different vowel sounds at different pitches to easily and efficiently move a cursor around the screen, and the same basic interface system can also be used to control a robotic arm:

This type of control is much more efficient than saying “arm left” and “arm up” and so forth, because the user is able to blend sounds together to generate complex, multi-axis movements. It’s still very much a work in progress, but ultimately, researchers are hoping that this technology will be able to operate all kinds of electromechanical instruments, from robotic arms to wireless home automation devices to electric wheelchairs.

[ Vocal Joystick ] VIA [ OhGizmo ]

Back in the day, dinosaurs were pretty awesome. Now, they’re pretty extinct, but the 160-odd million years that they were around gave them some time to study up on things like flight dynamics. Researchers are attempting to create a versatile unmanned aerial vehicle based on the body of a pterodactyl, specifically an early Cretaceous Tapejara wellnhoferi that used to live in what is now Brazil. The dino had a wingspan of up to 6 meters and probably weighed up to 15 kilos, which is a lot of potential payload. The idea is to recreate the pterosaur on not just a structural level, but actually mimic its entire biology, including flight musculature to enable to robot to fly just like a real pterodactyl would have.

It’s not just flight, though. Somehow paleontologists have figured out (or guessed) that pterodactyls were able to walk quadrupedally, run bipedally, and even sail, using their wings as sails and their head crests as rudders to steer. There’s a lot of adaptability there, and Petrodrones will have the ability to “alter their wing shapes using morphing techniques to squeeze through confined spaces, dive between buildings, zoom under overpasses, land on apartment balconies, or sail along the coastline.”

The prototype should have a wingspan of 80cm, and my guess is they’ll tackle the most obvious (and in some ways, easiest since it’s been done before) flying aspect first. Details will be presented October 7th at the Geological Society of America’s joint annual meeting in Houston.

See an artist’s rendition of what Tapejara wellnhoferi may have looked like, after the jump. (Read more…)

Finally finally finally finally FINALLY!!! I have been waiting for this moment ever since I saw Keepon in his video debut over a year ago. Since then, I’ve been posting Keepon vids whenever I can find them. I mean, how could you not desperately want a couple squishy yellow balls stuck together with two eyes and a cute little nose and an astonishing amount of dance talent (for having no arms and no legs and being only six inches tall)?

So yes, I’d sell my soul for Keepon. My soul, however, is not likely to be worth the roughly $30,000 that Keepon apparently costs. And that’s just the the base price, which I guess doesn’t include any of the little hats that he appears in. Unfortunately for fans like me, Keepon is designed primarily to do research on interactions between robots and children, so he’s expected to sell mostly to research institutions (and museums). But there is some good news… Keepon’s designers are “planning to come out with a new version which will use a simpler mechanism and have a cheaper price tag.” Significantly cheaper, I hope. And please, please, make sure it stays yellow and squishy with the awesome dance moves.

Keepon dancing, after the jump. (Read more…)

Getting robots to throw stuff like bowling balls and basketballs isn’t a walk in the park, but it’s much easier than the other end of things: getting robots to catch. Catching a moving object is (depending how sporty you are) either deceptively simple or deceptively difficult… You’re moving, the object is moving, and your brain has to figure out where the twain shall meet. Robots are good at math, but you need a lot of computer power and a gaggle of high speed cameras to figure out exactly where something like a ball is going to land.

So how do humans, who generally suck at doing complex math and playing sports at the same time, get it to work so well? For that matter, how do animals pull it off? Turns out that both humans and animals have just come up with a clever generic strategy that gets the job done, and Michael McBeath (an engineering professor at ASU) and Thomas Sugar (a psychology professor) have figured it out. Read how it works, after the jump. (Read more…)

Feeling ambitious? The Stanford Engineering Everywhere program is now offering a suite of computer and robotics courses online, for free. You can take the same three course introduction to computer science that imparts intelligence to most Stanford undergrads, and follow that up with Introduction to Robotics, Natural Language Processing, and Machine Learning.

The complete lecture portions of the courses are available on YouTube or iTunes, and you can download PDF versions of the handouts and assignments (answer keys included). What you can get online is identical to the courses taken by Stanford students, but no, you’re not gonna get credit for it.

[ SEE ] VIA [ Mr. Future ]

I’ve always appreciated the mathematical nature of the game of pool… How if you know trigonometry and physics and how to use those little diamond thingies along the sides of the table, practically any shot is possible. I know none of those things and prefer to play pool with my gut (in the figurative sense), but it seems like exactly the sort of game a robot would kick ass at. Deep Green is a gantry mounted 3 DOF industrial robot, with a wrist attachment that holds a pool cue. An electric motor is used to drive the cue during normal play, while a pneumatic CANNON takes over for breaks. The brains of the robot rely on cameras mounted both above the table and on the wrist joint itself, and it’s worth noting that there’s nothing special going on with the balls or the table; the computer determines which balls are which by looking at the different patterns and colors. So yes, you could set this thing up in your basement if you wanted some competition that won’t laugh at you.

The trickiest bit about this setup is the degree of repeatability and precision, but not accuracy, that the robot is capable of. Even sub-millimeter accuracy isn’t good enough for making shots across the table. Deep Green uses adaptive visual positioning to keep itself accurate to within half a degree, which is way better than my spastic technique is good for. As far as the game itself, much like its namesake Deep Green is able to look ahead and plan for multiple contingencies much faster and more efficiently than a human can. “Deep Green currently plays at a better-than-amateur level, planning and executing difficult combination and rail shots from across the table. It has pocketed runs of four consecutive balls, and it’s only a matter of time before it can consistently run the table.”

That’s right, only a matter of time. Sends chills down your spine, doesn’t it?

[ Computer.org ] VIA [ OhGizmo ]

Last year, we wrote about some R/C helicopters from Georgia Tech that were able to land on 60 degree slopes under computer control. Looks like Stanford has its own autonomous helicopter program, and they’ve been able to teach their helicopters to do all sorts of crazy stuff without anyone at the controls. The acrobatics in the following video are not preprogrammed maneuvers… A human pilot first performs some sample sequences, and a computer “watches” the trajectory of the helicopter and figures out how to duplicate (and improve upon) them. After the autonomous acrobatics in the video, you’ll see a set of six sample trajectories in color (flown by a puny human pilot) as well as a seventh trajectory (in white) which is what the computer calculates to be the ideal representation of the maneuvers:

There’s a ton of fancy math involved (read the paper here), but the upshot is that a computer can learn how to fly a helicopter better than an expert, after simply watching the expert fly for a while. According to Stanford, “in all cases, the autonomous helicopter’s performance exceeds that of our expert helicopter pilot’s demonstrations.” In of itself, this is not surprising, but the key here is that nobody has to program the robot to do anything specific. Programming robots to do what you want them to do is one of the biggest obstacles to practical robotics since it generally takes a substantial amount of knowledge and skill. Software like this, which allows robots to watch us and teach themselves, has a great deal of promise. And not just for helicopters… It’s also good for making omelettes.

You can see more vids on the Stanford Autonomous Helicopter Project’s YouTube channel.

[ Stanford Autonomous Helicopter ] VIA [ AI Robotics ]