robot bipeds (split from tutorial post - admin)

[Arislan]

I'd like to see some tutorial involving actuators, "muscles" and the like for more antropomorphic robots.
For example, I'd love to get some general info on creating a biped that actually climbs stairs!
I would think that would require sensors on the feet to detect the steps?

[Gopher]

First off, actuators is a broad term that includes traditional things like motors, servos, and solenoids as well as newer things like SMA's (such as Muscle Wire(tm)). Current SMAs are relatively weak, slow, and inaccurate, making them poorly suited to humanoid bots.

Bipeds are both very complicated and expensive. A biped that can climb stairs is even more complex. I may be wrong, but I think it was only fairly recently (about 5 years ago?) that one of the early versions of Asimo became the first humanoid bot to be able to effectively handle stairs at all. Of course, 5 years is a long time in the world of computers and robotics, but the point remains. You can buy humanoid bot kits, but as far as I know none are tall enough to even be physically capable of climbing a normal set of stairs, never mind the balance issues it raises. Well, you might make them crawl up stairs, but I don't think that's what you had in mind.

I'm not meaning to discourage you, but you need to understand that sophisticated humanoid bots are still the domain of the relatively hard-core, both in mechanical and software engineering. If you still want to tackle it, by all means go for it, but keep in mind this is not an exhaustively covered field, so you can't expect people to be able to walk you through the process. The people who could are busy building revolutionary new humanoid bots of their own.

[Arislan]

Gopher, thanks for the reply. This is what I had in mind:
http://www.ai.mit.edu/projects/leglab/robots/m2/m2.html

Unfortunately I don't see any detailed schematics on the site. Maybe because it wasn't finished. Their actuators look really cool, although the design appears to include a few redundant ones. They do have a photo of their actuator disassembled, I wonder where they got that motor.

[Admin]

as gopher said, you dont want to use artificial muscles for bipeds . . . use servos only!

search this forum for 'robo-one' and 'biped' . . . there are many other posts you might find useful.

ive only recently decided to build my own biped, and i probably wont start till next year, so i dont know much more than you . . .

[Arislan]

About the servos-only approach, have you any suggestion what type of motor or servo could be used for moving a 50kg biped around? I am talking about something like the ASIMO. I don't think the little RC servos can move its limbs.
I would love to scale up something like the robonova.

[Gopher]

Arislan:  A bot the size of asimo is likely to cost you quite a bit of money... are you sure you've got the funding? Honda spends a fortune on Asimo... they aren't even selling them yet, but a few have been leased to japanese companies for $150,000 a year. (yes, that's US$, not yen) Just want to be clear, I'm not talking about a few hundred dollars here; the kinds of servos and parts you'll need will be much harder to find by scavenging than parts for microbots, so you'll likely have to buy most of the components new. For the kind of strength you'll need at that scale, structural considerations are much more real as well. I haven't researched the parts, but I wouldn't be surprised if you were looking at at least $3,000 to put one together (someone correct me if I'm too low or high here). Again, not wanting to discourage you; if you can afford it, by all means, have at it!

Still reading? Servos. I've never looked into servos or motors that size, but they're probably intended for industrial applications, and won't be cheap. The combination of strength and responsiveness required to maintain balance will severely limit your options on actuators, so I'd start your research there. Then calculate how much power you need based on the specs for the servos (you'll need at least 12 - ankles, knees, hips, waist, neck, shoulders, elbows - and that's for a minimally-articulated humanoid with club-hands), and select your batteries. Then you calculate the weight of all that together, and do some math to verify that the motors are, in fact, powerful enough to move at least twice the weight of the servos and batteries. (you'll have the frame, the brain, and any sensors as well in the final bot).

When you find servos and batteries that will meet the bot's needs, then you can start building it; you could probably improvise something from cheap or scrap metal, but you might end up needing a few custom machined parts.

Once you've built the bot, then you get to face the hardest part of all: the software. It's not like you can just buy a biped brain to plug into your humanoid chassis, you'll have to develop your own. I won't sugar-coat it: unless you're an experienced software engineer, this will be completely beyond your abilities. I've been programming for more than a decade (not in robotics, but I've done some image processing, physics modeling, and dynamic kinematics in pure software) and while I'm sure I could accomplish it eventually, it would be a massive project and might take me years just to have the bot able to walk around, process camera data and identify things like stairs. You can't climb stairs by touch-sensors alone, you have know they are there, and be able to correctly align yourself to them. And a humanoid bot can only keep it's balance in two ways: the easy way is moving only statically (predefined movements, like the robowrestling competitions), but the hard way is dynamically (like Asimo.) You could statically script a bot to climb one particular flight of stairs, but not to handle stairs in general given variations in height and depth of the stairs.
 
Honda has spent billions of dollars and had a team of some of the best roboticists in the world working on it for 20 years, and they're still not ready to release a commercial model. So be warned, where you're wanting to go is largely uncharted territory, and as the old map legends used to say "here there be dragons!"

If none of that scared you into choosing a less ambitious project, then please, keep us updated as to your progress. We'll try to give insights where we can!

[Arislan]

Hi Gopher, it's very nice to hear what you are telling me. I like realistic appraisals. So that $3000 you mentioned is what you envision for each servo??

The difference for us is that there has been much progress in the last decades as far as robotics go. I am part of an R2-D2 builders club and looking at some photos you can see that back in 1976 some actual robots were already being built. At the astromech.net site you can see a few photos of film used droids from the time, they were filled with electronics and were actual programmable robots. I bet nowadays a microcontroller would replace the bulk of that wiring!

Anyway. I myself am no programmer of AI but I do know C++ and whatever I don't know, I have friends who do. My cousin is willing to help me as well. He's graduate student in electronics engineering. I am thinking of this robot, but also would like to keep it simpler than ASIMO. My design has claw hands with limited DoF.

My preoccupation is with how to get it to move. Mostly people say to use synthetic muscles, either pneumatic or hydraulic, but most of the instructions I find are for servo based bots. So I was thinking of simply scaling up a robotic biped. If I go for servos, the trick will be finding actual servos capable of doing what I need. And I'm not sure a regular motor can be converted into a servo...can it? If it could there are several options out there, like those used by electric grinders for example. Just thinking. Lighter parts could be moved by electric screwdriver motors and the like....Nowadays, tiny video cameras are not expensive at all. Two could serve as eyes. As for batteries, there are marine/RV batteries from solar power companies that I've seen before. Some are heavy but others not really that much. I've also been thinking about the modern hydrogen fuel cells which are supposedly not that expensive. The main thing is servos.

This thing about the stairs, though, there could be a system that informed the robot of which stairs it was climbing. Here is my crazy idea. In my house there is a stairwell, but also some steps in other areas. So, I could label the stairs with bar code stickers or similar. Upon finding stairs, the robot would scan for the bar code. If the code identified it as the large stairs, it would walk up the correct number of steps for it, if it was just the two steps, it would read that and climb that much. A more simple version would use photoresistors to read a coloured sticker. Red for the large stairs, green for the short, for instance.
Now, for dynamic balancing, I've been looking into what the options are for inclination sensors. I found one some that was easy to build, and seems cheap. It involved using glycol into a sealed cylinder. I have to find that link again.

So there are ideas out there, I'd love to find a way of piercing them all together, and your input is very welcome.

[Admin]

I say this like everyday, but:
As the size of your robot increases linearly, the cost and complexity of your robot increases exponentially.

Motors can be converted to servos using encoders. But for 20-ish joints in a biped, this will probably be way more trouble than its worth . . .

Have you considered using a Sharp IR rangefinder for detecting the stairs?

[Gopher]

I was stating 3K as an absolute minimum for the bot, though some of the major load-bearing servos (hips, knees, and shoulders are the main ones I think) might well cost thousands apeice for all I know, as I said I've never researched that kind of thing before.

As for programming, if you're willing to accept the limitation of having it able to traverse only a particular limited environment such as your home, then you can provide it with an accurate model/map of that environment in advance. IR rangefinders could be used to locate and position itself within this map, assuming it knows where it is turned on and retains an awareness of it's current location. Finding the stairs, doors, etc. and positioning itself properly would be much easier with the rangefinders than with cameras. Cameras may be cheap now, but real-time pattern detection and object recognition are still non-trivial programming tasks. It'd be interesting to see what you could do with large numbers of IR rangefinders; in the feet, for measuring and aligning with stairs, in the hands for aligning to grasp objects; it's one advantage of a larger humanoid, you have room to really cram it with sensors.

[Arislan]

Admin, I was indeed planning on using some IR sensors for distance readings. As for the servos, have you any specific tutorials or links that show how to make a servo? Someone gave me the name of a Bosch motor I might look into, but I haven't found it for sale in my area. It was said to be a Bosch GPA-750 and it is a scooter motor. I was told each motor is around $200 new.

Gopher, I did think of using a home base for the robot. The idea came from observing the Omnibot 2000 and its cardboard homebase plate.

I'm currently doing some vastly generic drawings of how I would want the robot to look. I'll post them once I get something a little more advanced. At the moment I just have very basic diagrams of limbs etc..

I plan on casting some exostructural parts in fiberglass, like the torso box.   

[Arislan]

This is an overly simplistic sketch of how I imagine the robot would be proportioned. I havent marked what some of those things are, but the "button" on the shoulders indicate inclinometers. There is also one on the head. The row above the eyes are a small sensor array. The eyes would be tiny video cameras. There is a grill on the mouth covering speakers. This is in case I decide to add a speech synthetiser.

Most logic circuitry would be located in the chest. Also the little boxes with the lightning symbol are batteries or fuel cells. The limbs are moved by servos of varying sizes. I think the legs and shoulders would have the largest motors. The robot has simplified claw hands. The arms are simply larger version of standard robot arms.

The only real oddity is the neck movement. I havent decided how to orchestrate that, but designing the head movements would be a challenge in itself....

Now to turn this simplistic idea into an actual robot, I would need to gather info on robot limbs and sizes of DC motors available. Then I could begin a more serious design of its skeleton, one that I could fit with the servos and motors.
Once the essential kinetics were designed, the more complex aspects could begin development.

[Admin]

As for the servos, have you any specific tutorials or links that show how to make a servo?

Nope, but if you put a encoder on a DC motor, use a timer and interrupts with a feedback PID control system, then you get a servo.
Oh, and also consider using a stepper instead. Might be easier . . .

Anyway, remember to keep your design as simple as possible. You might want to start your project just building a single leg. Then after running tests on the completed leg you can get a much better idea of what your robot is capable of (power requirements, control requirements, lifting ability, speed, etc.) I am not sure how many bots you have built, but if this is your first, your asking for a lot of headache . . . I would say this is very ambitious even for someone experienced . . .

[Arislan]

My only difficulty is getting these parts in Brazil. This place truly sucks for finding anything considered "non-essential" to everyday life. I hate it here. But it's cheap livin'...that's the tradeoff.

I agree, and it was always my plan to start with a limb and go from there. Although I'll probably do an arm first instead of a leg. 

I've noticed that in smaller robots the limbs are usually just brackets with servos in them. But in something as large as my robot, I think it would require actual "bones" in the structural design.

[Admin]

Yea I know what you mean, as Im considering moving to Thailand. Way cheap over there too, but technology is 5 years behind . . . expensive to get robot parts shipped over there too . . .

There is this huge place called Baan Maaw which is a perfect place to find scrapped electronic parts for anything you want super cheap . . . but again its all 5-10 years behind in technology . . . sigh . . .

[Arislan]

I am considering giving up making a robot until I'm back in America. This country really sucks. There is no way to get anything here without paying a lot of import taxes, bribes, and some stuff that gets ordered simply disappears!