Let's Make Robots!

robotizing a mechanical walking frame


Hello,

I'm looking for a device that records my walking motions digitally, and can control servos to imitate my walking forces, balance etc..

You see, I've got this 'mechanical spring-loaded bipedal walking frame' that assists in the walking motion (google search "spring walker" to understand what I mean), and what I want to do with it is turn it into a autonomous walking robot.

I've been thinking that if I attached servos and a computer (brain) to it to record my walking motions, then I can get it to walk on its own...without me strapped into it, that is. But I need advice on what kind of servos (make, model) to use that's powerful enough to move the walking frame. And, also what kind of device or program would allow me to record and save my actions in such a manner that I don't have to keep sorting out the data to tell the computer "when this, do this, etc (if p then q)? Presumably there's a device out there that would do all this for me, and not too expensive? 

I figured I could utilize the info gained by IR sensors taped to the frame to help with the decision making of what king of motion to do at a given situation (i.e. that it doesn't try to clime a hill when walking on the road, or walking when it's suppost to clime, etc.) Can you advise on what IR sensors are good for this particular job?

I don't know the scale of this kind of project, or if it's too difficult to do without expert roboticists on hand, but I'd like to try anyway.

Thanks very much, hope you guys can help.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

this is cool.

There seems to be a bit of a misunderstanding here about how the frame is balanced.  As a human uses the frame, shifts in weight are what keeps it upright.  Basically, this is a version of the inverted pendulum problem, which should be convered in detail in any text on nonlinear controls.  In addition to the leg servos, you'll need to mount some sort of weight on an actuated arm above the frame and use that to shift the center of gravity of the robot.  Search for information on self balancing (segway-type) robots for basics on accomplishing this.  Also keep in mind that bipedal walking is a process of falling foward and then moving the legs under the center of gravity to arrest the fall.  Working on balancing first is a step along the way.

 Just don't underestimate how complex your body's self-balancing ability is - you basically need to replicate the full system, including sensors (eyes, inner ears) and actuators (muscles) along with the 'brain' to tie it all together.

 

Good luck!

I’ve been thinking about scanning the walking frame to 3D (virtual) so that I can manipulate it easier.  What I mean by this is; if it’s in virtual, I can animate it and experiment with different motor strengths to weight (of the frame) ratio, change the lengths of leg, tension of springs, or even change the materials from metal to plastic (or even wood)! It might save time with all the physical experimentation and it may stop any physical damage coming to the frame, if I goof-up the work.

It’s just a thought, but any suggestions on this? Anybody have experience regarding 3D scanning and animation tools? Do you know of anywhere that can provide 3D scanning or reverse engineering? Thanks very much :)

Thanks for that Krowbar!

 Erm,…I’ve read somewhere stepper motors do not generate much torque, and as a result, slipping can occur (i.e. where the motor turns when no pulses are given or the motor fails to turn in response to the pulses). Also, that the control electronics are somewhat more complicated. And, in practice stepper motors are a good choice for precise, low torque motions, since they eliminate the need for feedback control.

 

I was thinking, therefore, maybe a permanent magnet brushed DC gearhead motor would be better than a stepper motor, since the advantages of DC motors include: simple to drive, cheap, and relatively high power-to-weight ratio. A disadvantage is that I’d need a sensor, such as an encoder or tachometer, to keep track of the position or velocity of the motor.

 

With a stepper motor, on the other hand, we can simply send a series of pulses to the motor, and it will rotate an angle proportional to the number of pulses. In theory, if we know how many pulses have been sent, we know where the motor is, and we know the velocity of the motor is proportional to the pulse frequency.

 

Oh no, this is a dilemma! Can anybody give more info & advice on this please?

My experience is that the amount of money you spend on a motor/gearbox is directly proportional to torque it produces. I don't think there's such a relationship between steppers and servos. For your frame either would need a big gearbox.

NB - unusually, that paper is writen in a nice, easy to understand format. There are plenty of pictures, too.

A servo might be a little underpowered, you might find better strength in stepper motors. A IR encoder could also be used with the stepper motor to provide input and stepper motors 'lock' when not told to do something.
Eric D. Vaughan, Ezequiel Di Paolo, and Inman R. Harvey at the Centre for Computational Neuroscience and Robotics, University of Sussex, have published a great paper titled “The Evolution of Control and Adaptation in a 3D Powered Passive Dynamic Walker”.

Although the title sounds imposing, the paper walks (no pun intended) you through all the modelling and design challenges involved with designing a bipedal robot capable of walking just like a human being!

I haven’t started reading it yet but it looks tasty!

see:http://www.informatics.sussex.ac.uk/users/ezequiel/evaughan_alife9.pdf

 

I’ve only just finished reading that tutorial. It’s really good. Thanks for the link.

Yeah, I was thinking a high level of abstraction is needed regarding translating my movements into “when this, do this” scenarios that are sufficiently removed from the particular, but not that removed as to render the robot incapable of recognising the situation altogether (such as, when there's a hill, it knows it’s a hill regardless of how steep it is). Erm…this is looking more and more like a lot of code-work, and I think, goes away from my original project intention   (i.e. to turn a walking frame into a robot, and teach it to walk using raw data gleaned from my ‘acts of motion’ when I walk, and apply that data to similar situations, so it (the robot) can walk on its own).

 

I’m trying to figure out a short-cut to doing this…I figured if I have the mechanical walking legs, all I’d have to do is 1) strategically attach servos to the joints to give it enough power to move on its own. Then, having recorded my movements as pre-programmes in a reader/processor (something like the reader from a motion-control rig for a camera) I’d 2) get the processor to control the servos to ‘enact’ my movements at the right time (i.e. when on a road, walk on the road, etc,.) I’m sure a device is already on the market that does all this. However, I might need recognition software (which is another thing altogether and probably very pricey).