PDA

View Full Version : Humanoid or Johnny 5 for research



musashi
01-10-2008, 05:22 PM
Hi again. As you know I'm a soon to be CS grad student.

I'm thinking of using a personal robot for some of my research. I've been thinking the Johnny 5. But maybe a Humanoid would be better?

Can I ask for any thoughts on this?

Here's my thoughts.

The Johnny has mobility. It seems that it can move a lot further and maybe this would be good for vision type projects that involve terrain?

The Humanoid may be good for research involving the dynamics of the human's movement?

I've been thinking, rather dreamy like, that a nice home beowulf cluster would be a great way to have tons of computational ability with a robot. What do you think of that?

As far as gripping goes, it seems both are fairly primitive right? Is the Johnny or the Humanoid expandable to include powerful gripping?

Thanks!

DresnerRobotics
01-10-2008, 06:26 PM
Here's my take on it:

Humanoids of any sort are going to be limited by weight and balance issues. Unless you are going with a very large, custom bot, having arms with enough torque and DOF to manipulate objects is going to be very difficult. Most the hobby sized humanoids are good for developing walking, dancing, fighting etc sequences... with varying control methods.. but sensory and autonomy is ultimately going to be limited by their payload, which is limited. Don't get me wrong, I can't wait for the wife to give me the nod to drop a grand or so on a new humanoid bot... but I don't plan on adding a ton to it.

The Johnny 5 model is not going to be as agile or flexible as a humanoid, but certainly more mobile by design. However, it has a ton of payload capability and a lot more room for expansion. For example, no way you'd ever implement a complete onboard PC on a hobby level humanoid bot, however this is easily done on the J5 model (wink wink, Project Page plug! :D ). I think Jim mentioned it has like a 6-7lb payload capacity?

However, on the note of gripping... while its impractical on the hobby humanoids (most dont even come with grippers) it's still not well implemented on the stock johnny 5 model. The biggest problem is that the arms are far too short. I am currently working on redesigning them to make them longer, and adding another DOF in the elbow, and once that work is done, he should effectively be able to pick up objects from the ground using 1 or 2 hands. The stock model's hands barely reach his treads.

If you have ANY questions regarding the johnny 5 model, feel free to ask, it's been my pet project for the past 3 months or so and I've spent more time planning and working on it then is healthy (or so my wife says)

SN96
01-10-2008, 06:34 PM
Absolutely, just what Tyberius said. Humanoids are good for motion dynamics, but you are limited to what you can stack on it.

Any rover, tracked or wheeled, has the ability to allow numerous devices, such as cameras of all sorts, PC boards like the nano or pico ITX boards, an array of sensors, and anything else you can think of. Rovers can handle a great deal of weight when compared to walkerbots and generally cheaper considering strong quality servos is the biggest part of the bill.

With a rover type platform you will have plenty of programming opportunities.

kdwyer
01-10-2008, 08:42 PM
Yeah, legs are great, but wheels/tracks are the way to go.
My Otto (J5 prototype) had arms long enough to reach the ground, but it also got in the way of the treads. His arms didn't have much strength, and the wiggle factor multiplied at each joint. I've pulled the arms and plan on going with extra sensors and WiFi. Flexibility!
A humanoid bot is gonna fundamentally remain... a humanoid bot.

One last thing:
Robots with tracks or wheels don't readily fall over.

JonHylands
01-11-2008, 06:47 AM
BrainBot is a Johnny-5 type robot, and having tracks definitely makes it simpler to do certain kinds of research.

Long arms are going to be an issue - when you make the arms longer, the shoulder servos are put under a lot more strain for lifting even simple items. We're going to be switching to using RX-64 servos in the shoulders, which provide a massive 800+ oz-in of torque. Hopefully that will make the arms more useful.

- Jon

DresnerRobotics
01-11-2008, 06:58 AM
BrainBot is a Johnny-5 type robot, and having tracks definitely makes it simpler to do certain kinds of research.

Long arms are going to be an issue - when you make the arms longer, the shoulder servos are put under a lot more strain for lifting even simple items. We're going to be switching to using RX-64 servos in the shoulders, which provide a massive 800+ oz-in of torque. Hopefully that will make the arms more useful.

- Jon

Yar, I'm going to be upgrading the two servos in each shoulder with 5955TG's when I do the elbow extensions. My wife is going to murder me.

JonHylands
01-11-2008, 07:30 AM
Note that my personal research robot, MicroRaptor, is a biped robot (not a humanoid though). I'm interested in walking, navigation, visual landmarking, and stuff like that.


- Jon

musashi
01-11-2008, 08:32 AM
I think you guys have given me some good information. Thank you!

I like the mobility of Johnny. I think it would allow me to do some interesting vision problems and maybe some AI problems too.

I think it's common to use the bluetooth control for the Johnny?

Recently I've started to get froggy and I was thinking of trying to build my own beowulf cluster for a brain for my robots. Any thoughts on that?

JonHylands
01-11-2008, 08:46 AM
Until you've written code that stretches the space and power available in a current high-spec PC (Core 2 Quad, 4 GB RAM), you're wasting your time with a cluster.

Don't introduce hardware complexities to a problem until you absolutely need to...

- Jon

musashi
01-11-2008, 08:54 AM
Good advice Jon. I do have a Core 2 Duo with 4GB. So you're right, it'll take me some time to eat through that.

Thanks again.

asbrandsson
01-12-2008, 12:38 AM
The Johnny 5 model is not going to be as agile or flexible as a humanoid, but certainly more mobile by design. However, it has a ton of payload capability and a lot more room for expansion. For example, no way you'd ever implement a complete onboard PC on a hobby level humanoid bot, however this is easily done on the J5 model (wink wink, Project Page plug! :D ). I think Jim mentioned it has like a 6-7lb payload capacity?

The problem that I find with the humanoid robot (I am talking about the robonova) is that it is not that useful in the stock form - meaning to make it useful you need to do a lot of work to it. Basically to the point where it is almost all custom.

Namely - double knees, grippers, sensors (including gyros, cameras, and tilt), a brain and a power system.

When you expand the legs to include the double knees, you can expand the body to take more weight, but if you want real payload you need to upgrade the leg servos. And with a bigger body you can up grade the arms, making them longer. Bauer mech has already done alot of this kind of work.

I, however, am convinced that you can put a pc on a custom robonova. I am trying to think of a way to built something around a pico-itx or a dual core mini-itx depending on what kind of weight I can get the legs to hold. Which I will have a better idea when I get the double knees installed on my robonova.

Asbrandsson

DresnerRobotics
01-12-2008, 01:07 AM
With mini itx formfactor you're still looking at a 7x7" board. I dont see how that would fit on a robonova...

Pico-itx is plausible by itself... but once you figure in the pico-PSU power supply (light, but adds a little size) and then the biggest problem, the battery, you're looking at a LOT of potential weight. With a 9.6v battery @ 4200mAh I get roughly 2 hours of runtime on the pico-itx using the 60w Pico-PSU. It also weighs nearly a pound, that is a TON for a humanoid.

With the weight and size of the combined pico-itx, pico-psu, HDD (even a solid state will add a little size) battery for the pico, controller for the robonova, and batteries for the robonova... I just don't see all that realistically fit into the torso while maintaining a proper center of gravity. I'm not trying to be a naysayer, I'm just trying to put things into perspective. If you want heavier duty onboard computing for a humanoid, I would look at going with a gumstix or Hammer.

asbrandsson
01-12-2008, 01:41 PM
Hello,

I am going to change to these servos

http://www.trossenrobotics.com/Hitec-Servo-HSR-5990TG.aspx

for the legs and arms of the robonova I have and I have to build a bigger body.

Don't get me wrong I have three er-1s and one of them I took apart and put the guts into a Hero body. They can take alot of weight, I put a computer, power, sensors and everything on it without consideration of stripping anything down, but I did mount a OQO 1+ on the front of the Robonova that I have and it could still move around.

With the Robonova on the other hand you have to strip everything down, even the power source will have to be a li-pol to keep the weight down.

Asbrandsson

DresnerRobotics
01-12-2008, 04:43 PM
Well, with a full replacement of 5990TG's, and perhaps a custom torso/pico-itx/lipos, it *might* be possible, however you'd still be looking at a tough scenario balancing the torso, as the pico-itx setup would add a considerable 'backpack'.

Personally, I would use the extra payload from upgrading to 5990TGs for setting up a wifi/camera link onboard, and offload the processing to a local PC. The benefit of an onboard PC just doesnt seem worth the hassle for a biped to me.

That said- more power to you if you're set on making this work.

kdwyer
01-12-2008, 08:06 PM
A tracked chassis would be simplest and cheapest, thereby allowing for an increase in size. That size increase allows for extra payload (read 'batteries'). It is precisely that excess power that is needed for onboard computers.
Unless you truly let the bot operate autonomously, ie. walk into the desert, turn it on, and leave for a week - you will probably be nearby, observing and monitoring (not gloating). You will also probably be near a WiFi or Bluetooth laptop or PDA.
You wanna get the most bang for your buck? Tracked vehicle, with wireless connection to a local PC. Simple, reliable, scalable, inexpensive. Humanoids are great, but too much of their weight and power budget are given over to just being.
Tyberias thought an onboard PC wasn't worth the hassle on a biped.
For that matter, why not free up that same space on a non-biped? Leave the thinking to the computers. We already have the 'brain' our bots dream about, just a wireless link away.

Damn, I think I've just talked myself into more work...