Berkeley - The Future of Robotics (Sexy) | Hacker News Reader

Berkeley - The Future of Robotics (Sexy)

youtube.com

38 pointsrwaliany16 comments

Heard near a Berkeley restroom:

"So, are you inferring the same grasp point as I am?"

That video was part of a video competition sponsored by the robotics startup Willow Garage's founder. The Berkeley video won first place (and $5000). You can find the full listing of submitted videos (including the $3000 and $2000 second and third place contestants) on Hizook.com:

http://www.hizook.com/blog/2010/08/21/pr2-video-contest-resu...

I guess some entrepreneur can find some other use to commercialize this bot.

That's faintly disappointing to me - a human can turn a sock inside out nearly instantly and without some convenient pole. The image recognition stuff about figuring out where to grasp it seems cleverer but it's still a little offputting to me because it all seems so contrived.

I guess I shouldn't be so down on it - there must be some pretty clever stuff in there. The image recognition problem seems harder to solve to me than the dexterity one.

I'm not sure where it becomes sexy. I guess the pole is faintly phallic, but as far as I'm concerned for a robot to be sexy it would need to look like Cameron out of Sarah Connor Chronicles. If the Berkeley researchers managed that then the future would really be here.

I'd ditto the 'I'm not sure where it becomes sexy'. I suppose putting a sock on a pole is one step away from putting a condom on a penis, but it takes roughly 5-minutes to put a loosely elasticated sock onto a pole. It would probably take the better part of a week to put a condom on a penis.

So really, this wouldn't be in the least bit sexy or even sexually useful. Perhaps it'll find use treating hypersensitive premature ejaculators with a robo-fetish, but I sincerely doubt this is a step closer to the sex-bots that have been predicted forever.

Someone seriously needs to seek out female companionship, even if it's just an old playboy stuffed under the corner of their mattress, if they found that in anyway sexual.

The video is certainly suggestive, but I wouldn't call it sexy. To each his own I guess.

Imagine it 15x slower, I doubt it would even look suggestive anymore.

15x speedup. What is the bottleneck that it's taking > 15x longer for it to do this task than a human?

#1 I would say it's hand placement is significantly slowing it down. It's putting its hands up when it's not in use, which means excessive hand movements between usage (not only time inefficient but also a significant waste of electricity and a burden on its mechanical parts).

Aside from that, it's simply processing speed and motor control. Processing power would speed up detection, which would allow it to act faster. Similarly better motor controls would help, especially built in resistance detection so it can tell how much force it's applying.

Also it's doing it completely wrong. A human at worst flips the sock over one of their hands, and at best does a sort of tuck & roll on their thumbs and stretches.

It seems kind of asinine that researches are trying to replicate human tasks without replicating human dexterity to achieve it with reasonable speed.

Some people in my lab worked on laundry manipulation, and it's surprisingly hard. My personal guess is that it's the lack of touch sensors. You can fold socks with your eyes closed pretty quickly, but if you had to do it with visual clues alone (e.g. by using a pair of those Gopher tools http://www.amazon.com/Seen-Gopher-Pick-Retrieving-Tool/dp/B0... ) it would take you significantly longer.

Probably mostly that motors which are

    - small,
    - accurate,
    - fast, and
    - cheap

do not exist.

RC servos meet all of those requirements, wouldn't you agree?

No, I would not. RC servos cannot follow specifically-defined trajectories, otherwise known as splines (ie. be at encoder position X at time Y), as RC servos rely on very basic (analog) position feedback.

Some of the lowest-level PR2 controllers require the specification of splines to achieve accurate motion.

Servos are pretty much terrible for anything except maybe driving the wheels on a very small robot, provided that robot is also equipped with encoders on the wheels. Basically you're just applying power to the motor for some time and hoping it turns the right amount; unlike say a stepper motor, you have no built-in guarantee of how far it'll turn. You can make up for that with the encoders, but the other big problem with servos is that they're not very strong... all the ones I've ever used have plastic gears inside which love to strip.

Probably except accurate, although I should probably add the criterion "torqueful" since that robot looks heavy.

It's only a matter of time before one of these kills somebody.

Reply on news.ycombinator.com