Subject:
|
Re: Walkers
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Thu, 15 Feb 2001 17:12:25 GMT
|
Viewed:
|
671 times
|
| |
|
|
> > > From a robot which attempted to balance using 2 wheels I found...
> >
> > From stationary start falling forward...
> > detect pendulum swinging forward
> > accelerate to bring upright again
> > pedulum is pushed back by acceleration
> > detect 'upright', stop
> > fall over really, really hard!
Try rebuilding you robot with the wheels tilted. Instead of the having the
axel parallel with the ground (0 deg), tilt the wheel so the axel is is at
60+ deg. This moves the effective center of rotation upwards. Once you get
it above the CG you won't fall over again.
> I've seen a device demonstrated that consisted of a lever with a free-
> swinging pendulum connected one end - with a motor at the other end
> of the lever. There were rotation sensors at the motor and the fulcrum
> of the pendulum. (This wasn't a Lego device) The software was able to
> jiggle the motor about until both lever and pendulum were
> pointed straight up - and to balance them there indefinitely.
> You could even push the pendulum with quite a bit of force to
> try to knock it over and the system would catch it and rebalance
> the system within a small fraction of a second. The entire thing
> was completely hypnotic because it worked with more speed and
> precision than you'd think possible. It was the robotic equivelent
> of balancing a pencil point-down on the tip of your finger without
> moving your arm...and then letting someone try to knock it off your
> finger without losing it!
>
> This was done by emulating the physics of the system in software
> and applying the precise amount of force theoretically needed to
> exactly halt the pendulum in the vertical position. By recomputing
> this solution about 500 times a second and adjusting the speed of
> the motor accordingly, the system could have the pendulum appear
> to stop dead in the correct position.
You don't have to know the physics of the entire system to do this
correctly. The pencil is the only important part. You can use feedback
control and some rough guestimates to replace knowing alot about the open
loop responce of the pencil point positioner. Quite a few years ago a
friend and I did the inverted pendulum thing in two degrees of freedom using
an XY table. When neural networks became all the rage, he implemented a
neural network based controller that balanced the pendulum (I never really
understood how he accomplished the training). It was once again revived a
few years later as a fuzzy logic control demonstrator.
|
|
Message is in Reply To:
 | | Re: Walkers
|
| (...) Well, I think your software has to be able to predict (mathematically) what will happen to the robot and the pendulum as it accellerates the motor. You can't rely on the simple "if pendulum in middle all is well" thing. You need to know the (...) (24 years ago, 14-Feb-01, to lugnet.robotics)
|
3 Messages in This Thread:
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
