More robot vids

[Gopher]

Thought this forum needed more robot videos, so I dug up some.  Links are to google video.

a rather interesting bit of

robot wrestling.

A very sturdy

4-legged walker

In case anyone missed it, the new

Asimo

. They say it can run, but I must nit-pick; it doesn't appear that both feet ever leave the ground at the same time, so it's actually just speed-walking. Still pretty neat.

This one

is just a line follower, but I thought it was interesting that they used a GBA as a microcontroller. Not sure if it's following the line with the camera, or if there are additional sensors that just aren't apparent in the video; from experience, I know the GBA isn't that powerful, so if it's using the camera, it's fairly impressive programming.

Now for something completely different. Not really a robot by most peoples' standards, but very cool: an

exoskeleton

. I want one, to crush those who oppose me. Muhahaha!

And finally,

something from japan

that's more bizzarre than impressive

[Admin]

Anyone have designs on those robot wrestlers? I want I want!!!

Japanese are whack . . . but amazing bot builders. One day Ill learn the language, move there, and be whack too. 

[Gopher]

Yah, those were my favorite too. They look like variations of the humanoid 'bot I see in one of your sidebar google ads all the time, all servos (or more likely those new things, don't remember what they're called, they're like servos but also contain a plethora of sensors for detecting pressure, resistance, etc) They're being remote-controlled, though they are obviously using an abstracted control system. Programming is a whole different ball of wax, but the physical construction doesn't actually seem that complicated to me.

[Admin]

well its all a bunch of preprogrammed actions tweaked to perfection, thats the easy part (although a bit time consuming). no robot intelligence/sensors involved.

i just want the mechanical/electrical designs 

one of these days ill make my own fully autonomous walker . . .

[Gopher]

See, the reason I don't even concider making a humanoid bot at present is that I wouldn't be satisfied with just mechanically preprogrammed actions. If I made one, I'd want to try to engineer it using an accurate intermal 3d representation of the world, generated through sensory data, and I'd want it to use inverse kinematics to actualy solve body positions based on logical instructions (grab X, push Y, etc). That's an incredibly ambitious thing, and quite beyond my present confidence level in robotics. I've experimented with programming IK solvers for 3d software applications, and in my younger days I did a lot of low-level graphics programming, including textured triangle fillers in assembly language. I've never tried to write an IK solver in such a low-level language, but I'm sure it could be done, but it would require a pretty powerful processor.

I loved the GBA one, because I've done some hobby GBA programming. It's actually very easy system to program for, and fairly powerful. A DS might be even better, and it is very similar in it's core hardware. It runs on an enhanced version of the same processor, and has many suplimental systems including dual-channel audio (recording and playback simultaneously), handwriting recognition, and 3d graphics. Picture a humanoid bot about 2" tall, built with a DS for it's core; the small screen is the face, and the larger screen on it's chest. You might be able to trick the handwriting recognition system into reading text from a camera (which are available as accessories). Probably not, though; I think most handwriting systems use timed vector information as input, not a 2d image. The little screen could render a fairly simple animated 3d face, while the one on the chest shows sensor and internal status information. It would need to solve the IK system and test it with an accurate physics simulation before taking action, so it might sit and think for a while between actions, but it could draw what it was thinking on the screen: showing a wireframe animation when doing an IK solution, showing it's 3d maps of the environment when it's processing sensory info or pathfinding. A hugely complicated project, but I know quite a bit about game programming, and much of the software is essentially the same. The hardware is a different matter, and I'm a relative novice in electronics and robotics, but the wrestling bots are clearly sophisticated enough for it so I know it's possible.

[Admin]

Hmmm well I dont have anything against preprogrammed actions, as it is a very powerful tool to get 'intelligent' actions with less electronic hardware. But I hesitate to call it a robot unless it is autonomous, meaning no remote control . . .

The main reason why I havnt built a walker yet is:

• 10+ servos can get expensive
• battery technology is still too heavy and not enough lifetime for 10+ servos
• a microcontroller+supporting electronics needs to be made super small/light . . . a skill i dont have yet

[Gopher]

Well, just as soon as I win the lottery (extremely unlikely, since I never buy tickets), I'll damn the cost and build a humanoid bot that's up to the task.

The trick is to remember the square-cube rule; you can't get caught up in a cycle of bulking up your servos, which requires more power, leading to bigger batteries, which makes it heavier, requiring bigger servos... sometimes you can't win that battle, and you have to go smaller instead. This is why these tiny bots, while not as smart as Asimo, are actually (proportionally) quicker and more agile in their movements (though their small brains make their movements less graceful and coordinated) Unfortunately, making smaller components is beyond the ability of most hobby roboticists; we just have to wait for the steady march of technology.

[JesseWelling]

Seems like a good start to a distributed microcontoller setup...program them all up to be controlled on i2c from a gumstix(http://gumstix.com/) and give it a little bit of the smarts (PID or fuzzy control) so the gumstix can focus on highlevel Python, clisp or C++ AI section.

But alas wheeled robots are the easiest and most versatile for small bots.