How to make a bipedal robot walking (smoothly) ?

[Ichan]

Anyone? Share please...

I am struggling with it for many days, and the youtube video below break my heart hardly:

Biped robot walks just like a human being.

What is the secret???

-ichan

[hate]

I don't speak the language the video is in but I'm guessing neural networks, gyros, lots of miniature actuators and last but not the least probably years of work. I must say I am amazed with what I saw.

[Ichan]

I must say I am amazed with what I saw.

Yep, watch the authors other videos especially the robot which ride a bicycle. Whatta...

My biped now can not even walk more than two steps, always fall or change direction just like my son when he was 9 month old. Easier to make a child which can walk than a robot.. just feed him and wait  .

-ichan

[elcielo]

http://www.intechopen.com/books/mobile_robotics_moving_intelligence

[solutions]

In control theory, it's called the "inverted pendulum problem". If you understand that one, you're all set.

When you figure it out, come back and tell me/us. This was PhD stuff back in my day - we were challenged to solve it as a senior project - "no thanks". Now it's control system basics from what I know about it.

Posted on: March 25, 2014, 10:11:12 10:11 - Automerged

Yep, watch the authors other videos especially the robot which ride a bicycle. Whatta...

My biped now can not even walk more than two steps, always fall or change direction just like my son when he was 9 month old. Easier to make a child which can walk than a robot.. just feed him and wait  .

-ichan

You're conveniently forgetting the other half of the "feed him" that has to be done  (or, like many of us, have you tricked her into thinking you're not able to smell that vile product?).

You try walking around with a load like that in your pants...

Posted on: March 25, 2014, 10:14:34 10:14 - Automerged
Oh yeah - bipedal is boring.

THIS is a chick magnet, LOL

https://www.youtube.com/watch?v=-vVblGlIMgw

She'll want you around to kill it 

[Ichan]

I am not a scientist nor engineer of electronics, i am just a practitioner or even better opportunist of it  - attached a link to a book "Modeling, Simulation and Optimization of Bipedal Walking" which is too hard for my jaw to chew off.

My approach is empirical, by defining several walking steps similar with picture below, simulating it on Autodesk Inventor and then adjusting all the robot joints (20 of them) so the robot does not fall on each step. But when the whole walking steps executed a stable walking is very difficult to achieve 

Adding Gyro will be on the next step, but i believe some commercial robot can walk without using it. I wonder if someone have the experience and walking source code of the Darwin OP robot (look at the price) to share.

Oh yeah - bipedal is boring.
THIS is a chick magnet, LOL

No, as being a male humanoid my bipedal robot can flirt at girls... 

-ichan

[hate]

Without a gyro or some other kind of feedback mechanism, a humanoid will be quite prone to effects to unbalance itself. I guess wide surface area for the feet can make walking possible without feedback but I don't think that can be considered walking because you're not modeling the human walking correctly considering even humans use ears for balance.

[solutions]

Mr Practitioner,

Walking is an initiated and delayed fall, not mere steps

[Ichan]

Walking is an initiated and delayed fall, not mere steps

I am all ears, please explain.

-ichan

[solutions]

Learn from yourself.

To walk forward you have lean to fall forward initially. At that point, your legs are catching up to keep you from falling forward more. This is the inverted pendulum balancing problem which is the key control. If you want to stop walking, you have to speed up to get the CG behind the support point (foot) - a Segway also has an inverted pendulum controller.

You can "walk" by ensuring support under the CG at all times, but when you want to stop, inertia keeps the CG going, you faceplant. An inverted pendulum controller will lean back to achieve a resting point for the CG.

It'll be a lot harder for you to empirically derive the control system vs read up and understand the inverted pendulum, IMO.

Not easy.

[pickit2]

so that goes someway to explain why I fall over, on my way back from the pub..

[solutions]

Ichan,

Please don't build a gay walking-robot like this one: http://cdnapi.kaltura.com/index.php/extwidget/openGraph/wid/0_6evhu7uy

Notice it does not address the challenges of turning your head while walking

[Ichan]

I've been more than one month being a fool trying learn how to walk  .

My reasoning on empirical approach is all that complicated math calculation (imagine inverted pendulum implementation on 20 joints) will not be possibly done in real-time by the mcu carried by the small robot. So it must be calculated off-line generating the walking steps required for later executed by the robot mcu. I thought that i can found that walking steps empirically and bypassing those crazy math.

So far what i found is COG alone is not enough, there is other term ZMP (zero moment point) need to be considered. Other thing i found is without any kind of feedback the robot joints need to be very rigid - mine has some weakness on the ankle about 1 degree play which can move the COG significantly.

so that goes someway to explain why I fall over, on my way back from the pub..

The knee is weakened and the gyro is non functional by the vapour of beer... 

Please don't build a gay walking-robot like this one: http://cdnapi.kaltura.com/index.php/extwidget/openGraph/wid/0_6evhu7uy
Notice it does not address the challenges of turning your head while walking

Bent knee walking it is called, i do not understand why million dollars project can not walk with straightened knee walking, like the video on my first post.

Turning the head to the girls will not move the COG significantly, but flirting at them sure will.. 

-ichan

[solutions]

http://www.youtube.com/watch?v=w40e1u0T1yg

http://gazebosim.org/

[Ichan]

I am pretty sure this robot move and walk using predefined motion sequences, and not using any kind of sensor feedback.

I decided to go down to the cave again, building the second prototype.

-ichan

[hate]

By looking at the size of its feet, it doesn't seem to be using feedback. Its movements look to be way too much preconfigured too.

[solutions]

Hey Ichy,

I found this one in my travels -may be useful for your toy

-----------

Feedback Control of Dynamic Bipedal Robot Locomotion
by Eric R. Westervelt, Jessy W. Grizzle
English | 2007 | ISBN: 1420053728 | 528 pages | PDF | 5.85 MB

Bipedal locomotion is among the most difficult challenges in control engineering. Most books treat the subject from a quasi-static perspective, overlooking the hybrid nature of bipedal mechanics. Feedback Control of Dynamic Bipedal Robot Locomotion is the first book to present a comprehensive and mathematically sound treatment of feedback design for achieving stable, agile, and efficient locomotion in bipedal robots.

In this unique and groundbreaking treatise, expert authors lead you systematically through every step of the process, including:
• Mathematical modeling of walking and running gaits in planar robots
• Analysis of periodic orbits in hybrid systems
• Design and analysis of feedback systems for achieving stable periodic motions
• Algorithms for synthesizing feedback controllers
• Detailed simulation examples
• Experimental implementations on two bipedal test beds

The elegance of the authors' approach is evident in the marriage of control theory and mechanics, uniting control-based presentation and mathematical custom with a mechanics-based approach to the problem and computational rendering. Concrete examples and numerous illustrations complement and clarify the mathematical discussion. A supporting Web site offers links to videos of several experiments along with MATLAB® code for several of the models. This one-of-a-kind book builds a solid understanding of the theoretical and practical aspects of truly dynamic locomotion in planar bipedal robots.

[htrx93]

I think a lot of the methods for this sort of walking revolve around a concept of a zero moment point https://en.wikipedia.org/wiki/Zero_moment_point, might be a good place to start reading.