Author Topic: 4 legged robot learning to crawl (Read 16710 times)

Admin · « **on:** August 21, 2007, 11:26:01 AM »

Interesting little critter that does NOT use preprogrammed walking gates . . . apparently it has sensors to determine its legs, body momentum, etc. It then runs this info through a simulator to design a functional walking gait.

[youtube=425,350]ehno85yI-sA[/youtube]

HDL_CinC_Dragon · « **Reply #1 on:** August 21, 2007, 04:50:13 PM »

Thats almost creeoy looking lol

IMO they could have applied a better vessel for its software. Perhaps a 6 legged walker like an insect and legs that had another axis since it really on has 1 ATM (X axis I believe) where If it had all 3 im sure thing could outrun a person... take a little longer to learn its gaits but I mean thats natural since its working with 3 axis' rather than just one...

very impressive none-the-less

Admin · « **Reply #2 on:** August 21, 2007, 08:31:45 PM »

With my robot fish I have a complex gait problem as well. Basically for every joint you add, it computationally takes longer to compute by an order of magnitude.

For example, twice the joints would result in like ~20x required computation time.

HDL_CinC_Dragon · « **Reply #3 on:** August 21, 2007, 08:33:29 PM »

I guess the computation time would be exponentially related to the amount of joints since it has a lot more possibilities right?

Admin · « **Reply #4 on:** August 21, 2007, 08:39:10 PM »

Yeap. The reason is because all the joints are coupled, meaning they arent fully independent. Modifying a single joint motion could change the walking gait of the entire robot.

Rebelgium · « **Reply #5 on:** August 24, 2007, 05:18:14 PM »

yeah I've seen this robot in youtube before, you can find videos there with this robot learning to walk with one limb taken off!

And you can also find a self generated video of the robot as how it "sees" itself through the sensors. You can actually see it's self generated self-conciousness evolve!

very very impressive and interesting

HDL_CinC_Dragon · « **Reply #6 on:** August 24, 2007, 05:22:41 PM »

you apparently didnt watch the whole vid didja?

Everything you said is already in that video lol

Rebelgium · « **Reply #7 on:** August 24, 2007, 05:28:44 PM »

I did watch it completely

What I'm saying is that there are more videos on youtube, one where you can see an image that the robot generated itself, and another one where that robot had a limb taken off and it learned to walk anyway.

HDL_CinC_Dragon · « **Reply #8 on:** August 24, 2007, 05:32:56 PM »

... maybe it didnt completely load for you? because it showed exactly what your describing.

Look at 3:03 minutes left and 1:50 minutes left...

Rebelgium · « **Reply #9 on:** August 24, 2007, 05:37:06 PM »

Oops you were right

Hhmm I guess I stopped watching when the robot went laying back on the ground again... $:-\$

HDL_CinC_Dragon · « **Reply #10 on:** August 24, 2007, 05:42:10 PM »

heh no worries

simple mistake =)

Robotboy86 · « **Reply #11 on:** August 24, 2007, 11:51:07 PM »

I always had thoughts about this kind of design my head. Taking a large amount of sensors(encoders for position, acceleramators, gyro's, even outside refrence points) you could build a system that would make it essentially have random movements, log in which each command did what and then compound them. IE if you have 4 legs with 3 joints, have it log what each one does when all the other ones are nuetral, and then have it log what it does with each at a given set.

This would naturally take a *super* long amount of time.. but it'd be a step towards "learning robots" that could adapt to situations without setting different scenerio's up. The real problem would be compression the data log's into something simplier..

Naturally you could probaly do something like vectors and reference them in turn with motor positions and what have you.

This would take a whole crap load of data storage but could yield intresting results.

kinda intrested now lol.

HDL_CinC_Dragon · « **Reply #12 on:** August 25, 2007, 12:24:44 AM »

eh get one of those 2 gig RAM sticks and use the memory from that to store everything

Rebelgium · « **Reply #13 on:** August 25, 2007, 10:09:04 AM »

yes, memory can be purchased very cheap these days...
Just look at ebay etc.
As long as you choose to buy an outdated sort of memory (eg. RAM DDR 333MHz) you'll get away fairly cheap.

Nyx · « **Reply #14 on:** August 25, 2007, 03:23:38 PM »

As a computer scientist... I would say that the problem nowadays isn't really having robots learn to walk and all... We can do that easily... We already could 10 years ago. The real problem, in my opinion, is that the hardware available for robotics right now is pretty limited. Everything is crazy expensive, and even for a high price, you get imprecise actuators that are also slow and bulky. Looking at Honda's Asimo, you can clearly see how limited it's movement range is.

Rebelgium · « **Reply #15 on:** August 25, 2007, 03:52:08 PM »

Yes, as a hobbyist it does bedome expensive.
But these prices are nothing for a large company... If they want to make a precise and very powerfull robot they can.

True the Asimo is very limited, but that is only due to the fact it's a commercial robot, using the cheapest part possible, to be able to make as much profit as possible. It doesn't have as goal to be very precise etc.

Nyx · « **Reply #16 on:** August 25, 2007, 04:32:23 PM »

Quote from: Rebelgium on August 25, 2007, 03:52:08 PM

Yes, as a hobbyist it does bedome expensive.
But these prices are nothing for a large company... If they want to make a precise and very powerfull robot they can.

I think that most robotics research teams are underfunded. Large companies tend to invest in products that already have commercial success... So they don't invest lots in research for something that may or may not work well.

You also have to think that to obtain the "ideal" actuator for humanoid robots, an entire specialized team would need to be dedicated just to that, and the things would be expensive, until they become mass produced.

How much budget does the Asimo thing have anyways? A few millions? It isn't that much, if you consider just the pay for the scientists in the research team.

HDL_CinC_Dragon · « **Reply #17 on:** August 25, 2007, 04:36:40 PM »

eh im not very impressed with Asimo... dont really feel like elaborating lol but I dont think it would be very hard to build something better than Asimo.

Nyx · « **Reply #18 on:** August 25, 2007, 05:40:01 PM »

Quote from: HDL_CinC_Dragon on August 25, 2007, 04:36:40 PM

eh im not very impressed with Asimo... dont really feel like elaborating lol but I dont think it would be very hard to build something better than Asimo.

I fully agree

Admin · « **Reply #19 on:** August 25, 2007, 05:51:03 PM »

Quote

eh im not very impressed with Asimo... dont really feel like elaborating lol but I dont think it would be very hard to build something better than Asimo.

back in the day Asimo was the best . . . but imitators have done way better in the last 5 years . . .

Quote

But these prices are nothing for a large company... If they want to make a precise and very powerfull robot they can.

Robots cost nothing compared to the pay checks a company must give to a robot engineer

Rebelgium · « **Reply #20 on:** August 25, 2007, 06:07:22 PM »

Quote

Robots cost nothing compared to the pay checks a company must give to a robot engineer Wink

Let's hope so

anyway,

Quote

As a computer scientist... I would say that the problem nowadays isn't really having robots learn to walk and all... We can do that easily... We already could 10 years ago. The real problem, in my opinion, is that the hardware available for robotics right now is pretty limited. Everything is crazy expensive, and even for a high price, you get imprecise actuators that are also slow and bulky. Looking at Honda's Asimo, you can clearly see how limited it's movement range is.

I don't think Asimo is the right robot to reference to, as Admin said, it's outdated.
Companies that do need robots do fund their research alot.

Too bad some specific parts have to be so expensive for the hobbyist though.

S. Karim · « **Reply #21 on:** August 25, 2007, 07:42:26 PM »

Quote from: Admin on August 25, 2007, 05:51:03 PM

Robots cost nothing compared to the pay checks a company must give to a robot engineer

oh really? thats good news cause i hear robot engineers get 30K on average which is like...doo-doo teacher style salary.

HDL_CinC_Dragon · « **Reply #22 on:** August 25, 2007, 11:05:19 PM »

what?? no way!... robotics engineers HAVE to make more than that...... right??? i mean cmon! Robots!

awally88 · « **Reply #23 on:** August 26, 2007, 03:19:02 AM »

I hope to all hell we get more than that! otherwise I'm screwed! 5 years of uni for 30k a year doesnt sound that good. lol

Rebelgium · « **Reply #24 on:** August 26, 2007, 07:08:43 AM »

Quote

oh really? thats good news cause i hear robot engineers get 30K on average which is like...doo-doo teacher style salary.

Ofcourse wages are different for different types of "robot engineers"...
This title can be overrated and used by people who are just assembling the robots etc.
Kindof the same way they use "Floor manager" for cleaning lady.

But anyway 30K will be one of the lowest, I'm sure real robot engineers make way more.

HDL_CinC_Dragon · « **Reply #25 on:** August 26, 2007, 03:32:40 PM »

"Petroleum Distribution Technician" = The guy who pumps your gas at a gas station (For those areas that dont let you pump your own gas)

JesseWelling · « **Reply #26 on:** August 26, 2007, 08:17:08 PM »

I work in the Embedded Control / Automation Industry and I do a fair bit better than that, especially considering that I live in Iowa...
But since I work in embedded product development I have to deal with a bit more mundane/annoying stuff than pure AI researchers...

Admin · « **Reply #27 on:** August 27, 2007, 04:53:45 AM »

Straight out of college, a robotics engineer can earn between $50k and $100k.

Just depends on how good you are and who you work for

But most of my robot making friends tend to start their own companies . . .

ed1380 · « **Reply #28 on:** August 27, 2007, 08:06:02 PM »

neato.

next thing these robots will learn is how to kill

HDL_CinC_Dragon · « **Reply #29 on:** August 27, 2007, 10:10:48 PM »

Well luckily theyre not smart enough to kill us all yet.

News:

Author Topic: 4 legged robot learning to crawl (Read 16710 times)