Video of dynamic standing

[StuartL]

It's not quite the amazing videos you see around the world but it's still something of an achievement.

This is my humanoid robot made from a Bioloid Comprehensive kit (+ two extra servos) dynamically standing. You can see that if you push him too hard he falls over right now but steps are afoot (badoom tish) to resolve this.

All software is scratch written, IMU input is from Jon Hylands bus IMU (http://www.huvrobotics.com).

Enjoy!

phpBB [media]

[JonHylands]

Very cool Stuart - nice to see people getting use out of those parts.

- Jon

[srobot]

Nice!

What did you program it in?

--srobot

[StuartL]

What did you program it in?

It's all scratch written C. He's doing full forward and inverse kinematics and feeding the IMU/gyro into the foot and ankle positions to cancel out his own movement.

More tuning is required but he's learned the basics

[tom_chang79]

StuartL, that is very impressive indeed... I haven't seen any dynamic ped control and inverse kinematic on the Bioloid until I saw yours...

It seems there is a hysteresis from the compensation... But this is impressive nevertheless... Keep us updated with more videos of your Bioloid!

Regarding your inverse kinematics, I'm assuming you are coding this in "C" or using AVR?

[limor]

very nice work!
have you looked at kalman filter to compensate for those oscillations?

[NovaOne]

It's not quite the amazing videos you see around the world but it's still something of an achievement.

You are too modest...It's a massive leap in biped control,.....one of a select few on robosavvy who has succeeded ..... Congratulations

[JonHylands]

As an interesting aside, one of my friends is going to be implementing a Kalman filter for my IMU this summer, and if it works, we're going to try and shoehorn it into the ATmega168 - if that works out, then the Bioloid IMU will become really useful.

- Jon

[billyzelsnack]

very nice work!
have you looked at kalman filter to compensate for those oscillations?

Do you mean put a second kalman filter on top of the IMU sensors' output? It looked more like feedback was the issue and he needs the gain adjusted in his algorithm.

[Will]

Hi from Will,
well done, i have not got to try my gyro thing yet, i'm still trying to understand my rx64 feet.
i've writen a behavior control which i've yet to try it is similar to
start
loop
load print id 120:25
if id 120:25 > 1658 then play motion page 50"lean forward"
else if id 120:25 < 1656 then play motion page 51"lean backward"
jump loop
end


does this sound like its going to stop my robot falling backwatds, or do i need to master programming in
C ?
i thought it might be as easy as "print ID 120:25" and then some if and then call. ?

again i say well done,
bye for now and be well from Will.

[StuartL]

As an interesting aside, one of my friends is going to be implementing a Kalman filter for my IMU this summer, and if it works, we're going to try and shoehorn it into the ATmega168 - if that works out, then the Bioloid IMU will become really useful.

We've got a working exponential smoothing algorithm that goes over the accelerometer and gyro ADC. We use different smoothing parameters for accelerometers to the gyros as the accelerometers are much noisier. The output of the exponential smoothing algorithms goes into a PID filter to control the inverse kinematics for the leg and ankle positions. I'm fairly sure these algorithms are both small enough to squeeze into ATMega8/168.

In fact we plan to rewrite the AX-12 firmware over the summer...

Do you mean put a second kalman filter on top of the IMU sensors' output? It looked more like feedback was the issue and he needs the gain adjusted in his algorithm.

I'm assured by my partner in crime (hereby known as RandomMatt) that a Kalman filter is a linear filter and doesn't work (well) for the non-linear space of 3D kinematics. He's a maths whizz so I trust him on this statement. The exponential smoothing and PID control are all children of his work and are both very compact and exceedingly fast.

You are 100% correct that the gain is a problem. It's slightly more complicated with the two layers of filters (exponential smoothing + PID) with four 'gain's to adjust but that's the core of the problem.

It's looking less and less likely that he's going to be walking by Saturday, which depresses me a little, but you'll certainly be able to see the dynamic standing in action.

[StuartL]

New video!

Note how the book collapses under his weight and with a short panic he corrects it...

phpBB [media]

However we've now completely broken the software with an attempt to make his feet adopt the angle of the surface he's standing on. Fortunately we have version control.

[JonHylands]

I'm assured by my partner in crime (hereby known as RandomMatt) that a Kalman filter is a linear filter and doesn't work (well) for the non-linear space of 3D kinematics.

While that is entirely true of a regular Kalman filter, there are extended Kalman filters that work very well in this space. Look at all the helicopter-based UAVs out there - they all stabilize themselves with a six-axis IMU and an extended Kalman filter.

- Jon

[Will]

Wow, thats cool.
and the spring effect of the book cover is an interesting test. well done.
My mom said my hoverbike was fantastic but man your robot seems pritty fantastic to me.
The yellow thing i got from tribotix a while ago in Data's chest might be one of those, man! i hope i get to learn about it.
i just watched your vid again, i'm trying to work out how to say "co o ol"
Bye for now and be well from Will.

[billyzelsnack]

Ah cool. I didn't notice before, but you stretched him too. I stretched my guy awhile back but I never got around to actually doing anything with him after the surgery. Are you having any problems with the servos not being powerful enough?