🚀 go-pugleaf

In article <1992Nov6.165554.16576@PacBell.COM>, tlhouns@ns.pacbell.com (Lee Hounshell) writes:
|> I had an idea for constructing one using "bend" sensors with weights attached
|> to one end.  As the robot's orientation would change, the sensors would move
|> (because of gravity pulling the weights).  If several of these sensors were
|> placed strategically inside a robot's chassis, I suspect that one could calculate
|> the robot's orientation fairly accurately.  Any comments?


Ah!  But wait!  What are the weights really telling you?  They don't hang
down with gravity, they hang in the direction of their net acceleration,
which is the classic problem with "cheap" inertial sensors.  Unless your
robot only samples the sensor when it's been standing still for a while
(which you need some other sensor to tell you), you don't kow what your
robot's orientation is, unless you can integrate your sensor perfectly over
time, and it doesn't have any mechanical error.  

This is a particular problem with things like the robotic helicopter task,
where the helicopter is basically depending in all six axis on inertial data.
The trend is to use cameras and stuff to look at the ground/horizon/etc. to
recalibrate a few time/sec.

So you've basically rediscovered inertial nav here, where your "accuracy" is
basically log($$$)... in the general case.

-Olaf