DPRG List  

[DPRG] two wheel balance

Subject: [DPRG] two wheel balance
From: Karim Virani karim at yadallas.org
Date: Fri Feb 22 16:06:54 CST 2002

Great work!

>>Now on to the inertial gravity reference tilt sensor.  Anybody know
>>anything about Kalman filters?  :>)

I read that the Segway uses a network of 5 "gyroscopes".  I saw a photo of
this network, and they were of course all arrayed at various 3D orientations
to each other.  They were all identical and also looked a lot more like
accelerometers than gyros I'm familiar with.  Is there any reason to assume
that the Where-Is-Up problem can be magically solved with 5 accelerometers?


-----Original Message-----
>From: dprglist-admin at dprg.org [mailto:dprglist-admin at dprg.org]On Behalf
Of David P. Anderson
Sent: Friday, February 22, 2002 2:54 PM
To: dprglist at dprg.org
Subject: [DPRG] two wheel balance


As Kip reported, the inverted pendulum balancing robot
as demonstrated at the last DPRG meeting is now balancing
on two wheels.

I've temporarily side-stepped the gyroscope-and-accelerometer
inertial reference sensor (but more on that to come) while I've
been developing the platform and control algorithms.  The 'bot
in it's current enlightenment senses it's tilt angle with an
aluminum "finger" that trails on the floor.

The previous three-wheeled version used a pot and ball-bearing pivot
mounted on the robot handle and connected to a pole and nerfball.
That was the setup demo'd at the club meeting:


It's important that the angle sensor have as little friction
as possible.  This one uses a very low resistance conductive
plastic pot and a replacement ball bearing from a high-speed
router.   Here's a close-up of that pendulum mount and angle


For the two-wheeled version the tail booms and caster wheel were
removed, and the battery centered over the platform axles.  The
3" main wheels were replaced with 5" wheels, and the angle sensor
was removed from the top of the robot and mounted under the bottom
of the platform with a short piece of aluminum rod contacting the
floor.   Here's a picture from the rear showing the angle sensor
mounted underneath the platform, balancing on two wheels:


Here's a shot of it balancing from the side.  Note the small table
it's sitting on, a measure of confidence in the balance mechanism!


The little aluminum disk mounted on the top can be replaced with
the ball and pole for demos, otherwise it serves as a handle.

The robot seems to perform much more robustly with this setup,
in several ways.  With the 3-wheeled version, when the stick/ball
fell forward, for example, the platform had to drive forward
to get in under the ball in order to achieve balance.  With the
2-wheel version in the same circumstance, as the wheels drive
forward they also torque the stick/ball backward and thus coming
to balance requires much less travel, wheel velocity, and
battery current consumption.  The whole thing seems more efficient,
and in the absence of any control signal sits "almost" perfectly
still, not oscillating to and fro as did the 3-wheeled version.

Similarly, when the 3-wheeled version ran into an obstacle it
invariably knocked the stick over and fell.  The 2-wheel version
simply bounces off, maintaining it's balance.  A sort of built-in
bumper behavior without any bumpers.

Lastly, I've hooked up a joystick and can drive the 2-wheel version
around by adding an offset to the tilt angle for forward and reverse
and added/subtracting a pwm bias to each wheel for steering, while
the robot itself maintains the balance.  Pretty fun!  I'll try to
make a video/mpeg of this when I get a chance.

Now on to the inertial gravity reference tilt sensor.  Anybody know
anything about Kalman filters?  :>)


DPRGlist mailing list
DPRGlist at dprg.org

More information about the DPRG mailing list