Robot Catches High Speed Objects 273
shpoffo writes "Engineers at the University of Tokyo in Japan have created a robot that can catch a ball moving faster than 186 miles per hour (300 kph) - more than 270 feet per second. It uses an array of photodetectors to directly control the three finger actuators - which can rotate 180 degrees in 0.1 seconds. It's only catching softballs at the moment, but operators are optimistic for it to soon catch other objects and grasp moving things. A video with odd sci-fi TV-series (coral cache) accents is available."
Sure, if it's thrown straight at it (Score:3, Interesting)
The robot is all thumbs. (Score:5, Interesting)
It would be even nicer if it had an arm to intercept balls that weren't thrown precisely to it though.
Is the US lagging behind Japan? (Score:5, Interesting)
But all the Robotic news seems to be coming out of Japan lately, is anything being done in the US that compares?
Note: Not asking because I think the US should be in the lead but that it should compete for the benefit of all, definitely the US had the first industrial robot back in 1962 AFAIK:
http://en.wikipedia.org/wiki/Industrial_robot [wikipedia.org]
And it's rather sad to think we're lagging in this on the R/D side in new frontiers. Unless this should be the extent of it:
http://robots.engadget.com/entry/0657766019921755
Fast controled motion robotic (Score:3, Interesting)
As mentioned, there is no arm and the area for interception is very tight. Building an arm mounted interceptor may raise serious problems with inertia though.
Time to think of a robotized pickpocket.
Define catching... (Score:5, Interesting)
Picking up an imprecise, reasonably fast throw to a particular area doesn't need catching ability : think of those coin collectors on toll gates which are just a funnel down to a small coin slot.
So it's really a display of fast reacting robotic actuators and a pretty cool photo detection in order to time the reaction correctly. As the guy quoted in the article says "It's an engineering feat really"
Real catching, in my opinion, can only be acheived if you can follow through with your hands to "take the speed off the ball" at least for hard objects. I think that a fast moving real baseball would be incredibly hard to catch robotically. A mitt is really useful because it allows the momentum to be absorbed into a wide area. In cricket, all fielders know they have to bring the ball in to their chest or follow its trajectory after catching impact to not lose the ball - they don't have a mitt. This robot couldn't catch a moving hardball no matter how fast its actuators are, because the kinetic energy has to be disspated properly, and with a heavy ball this energy is very high.
Pretty cool demo though. I think its applications will be rather more in the picking up of (reasonably slow) moving objects realm than any useful rôle in catching. If you want to catch soft balls all day long might as well just breed dogs.
Re:High speed moving objects?! (Score:2, Interesting)
I don't know how hard a softball (sic) is, but a cricket ball is solid cork wrapped in leather. And I have the bruises this morning to prove it, after playing at the weekend...
.
Re:High speed moving objects?! (Score:1, Interesting)
This device would help a lot Bush's bodyguards on a next G8 meeting.
Robocop 2 (Score:2, Interesting)
Cue that scene where Robocop catches a bullet fired to a police.
I've always wondered about the real physics of that scene, maybe robocop's fingers would be destroyed, or the bullet deformed... all that kinetic energy has to go somewhere...
Ok, back to work.
Re:Define catching... (Score:3, Interesting)
There's nothing that is fundamental about the way humans catch. We happen to use hand motion to absorb speed absent a glove, but all that's required to catch is that you absorb the energy somehow. A robot arm could do it just by being tough enough to take the hit.
"I think that a fast moving real baseball would be incredibly hard to catch robotically."
It would be hardER because it has no spongy give and would bounce off the palm of the hand if not caught precisely.
But it can be caught precisely.
Non-linear equations (Score:5, Interesting)
In the video, the ball has a very visible parabolic flight curve over the 2 meters distance.
The parabolic flight curve actually makes this a harder task. If the equations of motion were purely linear, then it would be a simple task to calculate future position. The second order nature of the trajectory mean that a little more maths is needed to predict where to catch it. Much of the maths for this sort of thing uses matrices (read linear algebra) which would fall over for this task.
I seem to recall that human cricketers use a simple technique for solving this problem. As they are running to catch the ball they move so the ball is kept at a constant angle in their field of view. Keeping this angle constant ensures that the ball will neatly arrive in their hands. Or so the theory goes.
I've long thought that catching a ball would be a great research project, mainly due to the quadratics calculations involved, great to see it realised.
Re:Shows just how powerfull the human brain is (Score:3, Interesting)
That's a robot? (Score:3, Interesting)
It seems to me that's pushing the definition of robot a bit much. It's a grabber that closes when something approaches it. The ball is thrown straight at it. It seems more like the doors at the supermarket that open when you approach. Of course, the doors won't open fast enough for people moving at 186 mph but it's the same general principle.
The impressive thing about all this is that I was able to download the 9+MB video, first try, using the link on Slashdot's front page, in about 15 seconds. Now that's technology!
Wake me when someone builds a working pusher robot [somethingawful.com]...don't bother me with this "hand robot" jibber jabber.
Re:Non-linear equations (Score:3, Interesting)
I should explain why this hard to solve in a machine learning context. Whilst I agree that quadratics eqns have long had explicit solutions this may not be relevant if you use a machine learning approach. For current AI the game is not to program in the explicit equations themselves, but more construct a system in which the machine can learn through trial and error, using a positive feedback mechanism such as neural networks. Most of the work I saw in this domain five years back, when I was working in the field, used a strictly linear approach so would fail here. It is an interesting computational challenge to have a system which can "learn" the equations of motion.
I mean, if you consider that a ballistic trajectory in a vacuum in a uniform gravitational field (such as experienced to a good approximation by a ball thrown across the surface of the Earth at reasonable speeds) is completely determined by the just three parameters (position, velocity, and gravitational acceleration)
But in reality we are not working in a vacuum, as well as these effects we also have viscosity, wind resistance, and wind to take into account. Strict adherence to a quadratic equation would fail every time. In some respects having a fast moving ball makes the computational question easier as there is less contribution for these effects, it does make it more of an engineering challenge though. I'd actually be more impressed by a slow moving ball with lots of spin on it over a greater distance.
Re:Sure, if it's thrown straight at it (Score:3, Interesting)
Neither is having the hand without the being able to move and position it with an arm. The hard part is moving the hand in position to catch the ball. I'm not terribly impressed by just the hand alone, especially since they're still only using soft balls, like foam rubber balls. They're not even softballs, which aren't really all that soft, by the way.