Subject:
|
Re: Detecting tilt with an Accelerometer.
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Mon, 21 Oct 2002 17:18:12 GMT
|
Original-From:
|
T. Alexander Popiel <popiel@wolfskeep.com/StopSpam/>
|
Viewed:
|
853 times
|
| |
|
|
In message: <20021021164325.11088.qmail@verizonmail.com>
"Steve Hassenplug" <hassenplug@mail.com> writes:
> I've had discussions lately with a VERY large number of people about using
> an accelerometer to detect if an object (like a LegWay) is tilting.
>
> I'd like to continue the discussion here, with anyone willing to talk about
> it.
>
> Let me start by saying, "It won't work."
>
> I could be wrong, and I'd be willing to listen to anyone who can prove
> otherwise, but I really don't think it will work.
Do you have some concrete reasons why it won't work, to go along with
your assertion (for those of us just now joining the conversation)?
I suspect that it could be made to work (in conjunction with other data
like what the motor output is, etc.), based on the fact that the human
sense of balance is based in large part on a set of four 2-dimensional
accelerometers (in pairs set at right angles). (Of course, our sense
of balance is augmented/calibrated by a horizon-finder... but it does
continue to function with an amazing precision and accuracy with eyes
closed.) Using accelerometers to determine tilt is of course dependent
on a consistent gravity field, but here on the surface of the planet,
I don't think that's too much of a problem.
The tricky part will of course be taking into account the expected
accelerations due to intended motion. That would be influenced by
the first derivative of the motor output over time combined with
attitude over time... you'd have to have a tiny physics simulation
going on to separate the acceleration components due to self and
due to external factors. Since geometry of the robot is fixed, the
simulation may be reducable to a small number of multi-dimensional
table lookups...
For a thought experiment, I'd propose a robot with two big wheels
attached to two motors, and 3 linear accelerometers set at right
angles. That's five elements of interest, with 1-dimensional
data from each.
To start with, I'd use the prior accelerometer readings and motor
outputs over a time span sufficient to account for sensor lag and
transient motor startup/slowdown effects (plus any applicable gear
lash) as the inputs to the simulation. Measurments every 5ms over
a period of about a second is probably sufficient... for a total
of about 1000 inputs to a closed-form inputs-to-attitude function.
If the robot is actively trying to maintain a vertical attitude,
then much of the domain of the function will never be used...
This can probably be significantly simplified by finding patterns
in the data, but that sort of analysis I leave until after I've
moved beyond the thought-experiment stage.
- Alex
|
|
Message has 1 Reply:
 | | Re: Detecting tilt with an Accelerometer.
|
| (...) Alex, I think that you would only require two accelerometers, one positioned low near the wheels, the other near the top. Both should be oriented to measure accelerations along the direction of travel. Using two accelerometers gives you a (...) (22 years ago, 22-Oct-02, to lugnet.robotics)
|
Message is in Reply To:
| | Detecting tilt with an Accelerometer.
|
| I've had discussions lately with a VERY large number of people about using an accelerometer to detect if an object (like a LegWay) is tilting. I'd like to continue the discussion here, with anyone willing to talk about it. Let me start by saying, (...) (22 years ago, 21-Oct-02, to lugnet.robotics)
|
14 Messages in This Thread:
 
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
This Message and its Replies on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
