On the Gripping Hand

eek_the_kat writes "The Sensor Fusion Project at Ishikawa Hashimoto Laboratory has developed a high speed visual feedback system called SPE-256. It allows the robot to track fast randomly moving objects and grasp them (movies here). The applications seem endless! I have seen many robot mpegs as of late, many courtesy of /., but these have to be some of the coolest I have ever come across. A must see."

Comments

I covered this last year

[Goon Number 1]

Here.. "It looks like high contrast items are needed for the tracking system to work optimally, but a combination of sonar overlayed with ccds and IR would likely make that less of an issue. It appears that vison is done through a video camera that tracks the moving object, and in turn controls the arm. Also, interesting reflex action with the thumb serving to close the hand once contact is made. Hmm, as I watch more of the videos I'm less and less impressed, It looks like the handler is actually all but feeding the objects to the arm, not unlike teaching a kid to catch a ball by placing it in their hands. That's a shame really." (Edited for spelling)

Is it real?

[NemesisStar]

First up, hooray for me I managed to get to the site before it got slashdotted.

Ok, now that I'm done with that, am I the only one that's dubious as to whether this is real or not? Admittedly I've only inspected the videos and not checked out the rest of the site, but it looks to me as though a lot of those moves can be programmed in and "acted" out by the object.

The best example of this is the handshake. Notice how the hand is not even in the right position to SHAKE a hand until near the end when the hand rotates 90 odd degrees (so it looks like a hand that could possibly be shaken) and then the human hand moves in nice and slowly. Any old dolt can shake a fake hand, I'll bet the robot hand doesn't give half as firm of a handshake as my little brother. :)

It even looks as though the robot shakes the hand one more time than the human expects.

I'm sure that they've made leaps and bounds into robotic hands, but I can't help but suspect that they're playing it up for more than it's worth.

Re:Is it real?

[Planesdragon]

How does the robot arm locate the object in a 3 dimensional space, using only one eye?

I'd be more impressed if they used two cameras to simulate "depth perception", myself. I have yet to hear of a setup that used stereoscopic vision.

As to your question, try covering one eye (you good eye, too!), picking up a spoon, and then trying to touch it to something in three-space. It's not as hard as some would have you believe, and I suspect it won't be as hard for a computer to pick it up, either.

Re:Is it real?

[zmooc]

When picking up a spoon, you can feel the object it's lying on. Try putting a pretty tall object on a flat surface and look at the top from the side on a distance with a stretched arm so you can just reach it. Now don't look at the flat surface but just at the top of the object (as if it was somewhere in the air without any point of reference to help you - like the robot). Try to fetch it. It's a lot harder now.

In the more advanced examples it did use stereoscopic vision. Look at the one where it repeatedly catches a ball. The same for the video where it catches a ball which falls vertically. Only in the tracking-examples it did use a single camera if I recall correctly.

Is it real?

[PhunkySchtuff]

It looks like the camera is doing some pretty funky location detectiion - in 3D.
I could only see one camera in their schematics and in all the videos.
How does the robot arm locate the object in a 3 dimensional space, using only one eye?
Other than that, it looks very cool...
-k

Teaching a new dog old tricks

[Anonymous Coward]

It looks just like a person playing games with a dog! My girlfriend's terrier plays like that, tracking a waved object, waiting until it slows down enough to be grabbable and then darting in. Wonder if the reserachers had to play with their robot to train it?

Really impressive, but...

[allanj]

I'd like to know how many times it fails before it manages to grasp the objects. If it fails, like, 50 times for each success, then I'm a lot less impressed. I saw the videos (server not slashdotted form where I sit), and the speed and precision with which the hand moves around is really impressive - sure hope this is for real.
I've been doing some robot control software myself (trying to make it drive towards a moving target, using vision guidance) and that much simpler task was hard enough.

Re:How fast

[Anonymous Coward]

Got a look at two of the clips.

I would describe the speed as on the order of 1/3 the speed of a dog being teased with a tennis ball.

And the sequence looked about the same; the robot hand follows the ball back and forth just like a dog a then lunges out to grab it when it sees its opportunity.
>
>

Another thing if you noticed, the motion of the ball on the stick *WAS NOT* random. It was basically back and forth.

You could kick the ball around and the dog would still track it.

Creepy hand

[spineboy]

I don't know about you guys, but this thing
CREEPED ME OUT! I got the willies just looking at it. The way it moves is so un-human, un-animal..

umm yah it's cool, but brrrrrrrr - gives me the shivers. Anyone else agree?