Tuesday, August 04, 2009

“What's Behind Japan's Love Affair with Robots?”?

Nothing says slow day at the Tokyo bureau like yet another story about Japan’s love affair with robots arrRRRRRRRGGGHHhhhh…. Seriously, I’m tempted to write a story (in Japanese) entitled The Western Media’s Love Affair with Japan’s Love Affair with Robots. If other Japanese follow suit, I can blog about Japan’s Love Affair with the Western Media’s Love Affa….
Every year or two, Japanese researchers roll out a new robotic invention — the latest to grab headlines earlier this month was a mechanized baseball duo of a batter and pitcher that can throw 90% of its pitches in the strike zone.
Never mind what an awful sentence this is—it’s the pitching robot, not the duo, that throws the ball—hasn’t the reporter seen a pitching machine in her life? The magic here, it should go without saying, vests itself in the robotic batter, who (which?) can put the pine on the ball with a monotonous precision that is capable of shaming Ichiro into premature retirement. If (and this is a big if), the batter robot is able to do this without any direct input of information from the program that controls the pitching robot—i.e. the pitching and batting systems are independent—it is a spectacular stunt not unlike a successful missile defense system test.
In the past several years, Japan has committed several tens of millions of dollars to an industry…
Does the writer think that the U.S. robotics industry hasn’t committed “tens of millions of dollars to” robotics? (R&D? production?) Bring a sense of proportion to you numbers.
“Robotics is to be for the Japanese economy in the 21st century what automobiles were in the 20th,” says Jennifer Robertson, a professor of anthropology at the University of Michigan.
It’s always possible, but I don’t think that an anthropologist living in Michigan is the best prophet when it comes to predicting Japan’s industrial structure in the 21st century.

I could go on. But it’s past midnight. And I’m soaked.

Mark, Matt, Michael, M, I’ll get back to you tomorrow. (All Ms? That’s weird, if only because the initials of three prettiest girl in my high school sophomore class were MM. And Marilyn Monroe was still fresh in our memories.) Hopefully. Sorry about that…



Someone should compile a list of these BREAK GLASS IN EMERGENCY aids for functional illiterates. Let’s see, Pocari Sweat (but not, interestingly, Calpis), porn magazines on commuter trains…

urk…

4 comments:

Jan Moren said...

Just to correct you a bit, but it's the pitcher that's the greater advance, not the batter (and I say that as someone who's been involved in another humanoid robot batter). The point of the pitcher is that it can pitch accurately and with speed given an arm-and-hand like configuration. The time window for a coordinated release of the ball in the right direction is very small, and doing stuff like this helps us understand how the biological brain can manage it.

Also, while overstated, the car analogy has a trivial basis in fact: the Japanese auto industry is the largest financier of robotics research (think Honda Asimo, for instance; Toyota recently pretty much bought up the entire robotics research division of Sony), likely because they see the car market writing on the wall and is looking for another product with the same kind of features as the car.

Jun Okumura said...

Janne:

Thanks! Yes, the pitching arm—I saw it on TV—was a beautiful piece of work. It must be an exquisite programming and mechanical engineering feat. But you’ll have to explain just how your batting robot functioned before I accept your take on the technological superiority of the pitcher. Till then, I’ll continue to believe that building a batting robot (with wrist and elbow action and all) that can determine the speed and trajectory of a incoming 140 KPH sphere that dips and weaves while missing the strike zone altogether 5% of the time and whack it is the greater feat, and that the computer code for the batter must be much longer and the sensing system incomparably greater than those for the pitcher.

With regard to Toyota’s interest in robotics, note that Toyota is a global company whose manufacturing and to a lesser extent R&D functions are increasingly moving abroad. Note also that Sony—another globally-oriented—company chose to sell it. It’s a wash as far as Japan is concerned. Thus, the story says something about Toyota and Sony, but little about Japan, which was the TIME correspondent’s point.

Jan Moren said...

Jun, here's the robot I was involved with (the batting was just one example task we had; I worked on the vision side, but the focus is on motor control): http://www.cns.atr.jp/icorp/topic2e.html

I don't want to go into this in any detail (and believe me, nobody wants hear all of it), but the pitcher really is the interesting part while the batter is not.

A game of physical skill, like baseball, is interesting because it highlights the limits of human performance. We find the game interesting because it's just on this side of the possible to do. All the parameters of the game - the distance between pitcher and batter; size of the ball; length and wight of the bat; size of strike zone and so on - are delicately balanced so that both pitcher and batter have a very difficult time, and neither has a decisive advantage. If pitchers started to easily strike out batters or if batters started hitting the ball all the time, the rules would soon change to restore the balance.

So we know that for humans, and human-like machines, the pitching and batting is a difficult but attainable goal. But that goes only for _human-like_ machines. That is, with force and reaction time roughly equal to humans, and with perceptual capabilities about the same as humans. The robot I linked to above is expressly built to mimic human capabilities. It needs to keep its balance while swinging a bat, and it needs to estimate the ball trajectory using only the two cameras mounted in the eyes. That is really difficult, and it pretty much sucks at hitting the ball.

Here's the Youtube clip of Ishikawa's baseball duo: http://www.youtube.com/watch?v=m0Fjkzc_gvw

The reason I'm unimpressed by the batter is batter is that it doesn't have those human-like restrictions. Specifically, look at the left and right side of it, and you'll see two big round cameras mounted on gimbal joints that track the ball. Those are high-speed cameras shooting at several hundred frames a second - much higher rate than human eyes - with their own light sources (the white round collars are strong IR lights). And they're set more than a meter apart, which is a baseline many times longer than the typical 16cm or so that humans have between their eyes. That makes estimating the ball position and speed quite easy.

As I said above, baseball batting is designed to be difficult for people like us, but as this robots ignores those relevant restrictions it makes the task much easier, and frankly no longer very interesting. It's kind of like if you started to pitch and bat using balloons instead of balls - much easier but it removes the point of the activity. It doesn't tell us anything about how humans do this task (the very fact that they opt for high-speed cameras and long baselines tells us they are not even trying to; humans rely only partially on disparity cues for this), and it doesn't create anything new from an engineering standpoint either - it's a long-solved engineering problem after all. It's a fun toy, but that's more or less it. And if you delve into his actual work you'll notice that the batting robot is an old, finished project.

Accurate pitching, on the other hand, _is_ interesting. You can't really do closed-loop control of the entire movement sequence - you dont have the time - so you need to do some combination of closed-loop and open-loop with off-line feedback. The robot essentially has to learn to pitch by practicing (whether for real or tuning it in simulation). In my old lab they have a similar project having that robot at the top throw darts. It's the same problem, and it's difficult. Yet another example would be playing the piano, where you don't have the time to get feedback on what key you pressed before you need to initiate the next movement in the sequence.

If you want details about this let me know, but I think this is getting a bit long already.

Jun Okumura said...

Janne:

The following is my understanding of pitching and batting robots based on your comment and the accompanying links:

There is no way to determine which of the two robots is more difficult to develop unless it is determined how closely the robots and external conditions conform to the real-world and what the measures of success are. From that point of view, the arm on Ishikawa’s pitching robot does appear to be closer to a human one than the one on his batting robot in that it appears to actively use its wrist and possibly fingers, while the latter appears to be a single-joint device. I’ll wager that developing a successful batting robot would be an exponentially more complicated engineering feat if, like human batters, it had to swing the bat with two identical arms, since it would have to use coordinate multiple joints—elbows and wrists, at least—to bring the bat through a plane level enough to have any realistic chance of hitting the ball. Could such a robot work with a purely closed-loop control system, or, like real-life batters, would it also have to use a combination of all three elements?

Note also that both robots, unlike your more humanoid experiment, are merely arms fixed to their respective mounts and designed to swivel around the point of fixture in a level plane. And of course the external conditions—distance between the pitcher, weight of the ball, etc., etc.—are unlike anything that pitchers and batters encounter in the real world, which fact makes you wonder about the meaning of the 95% strike rate.

Thanks, Janne. Once again, you’ve help me think better.