From: Tim Wescott on
Alfred Bovin wrote:
>> The IMU can _help_. In particular, the IMU plus a GPS receiver will
>> vastly increase the short-term accuracy of the GPS measurement. It won't
>> be good enough to resolve problems with multipath on the GPS, nor will it
>> carry you through a very long GPS dropout.
>> Odometry and visual data may help, but it's problematic -- how do you tell
>> the difference between a wheel slip and the sudden onset of GPS multipath?
>> In theory you could make a filter that would sort this out (web search on
>> "particle filter"); in practice you could spend a long time dinking with
>> such a filter without getting something that is robust and practical.
> Well the whole idea of using visual info was to be able to tell the
> difference between actual motions and wheel slips and also so I can get a
> full 3D motion reconstruction and not just the planar info from the
> odometry. The full 3D pose is important when "doing stuff" in rugged
> terrains.
> As I wrote in another reply, the visual system is rather slow and may fail
> sometimes so my thoughts came to using an IMU. What I don't know yet if
> whether this particular IMU is too noisy to make a combined improment on the
> visual and odometry data.
If the visual system can give you absolute measurements and not just
deltas then whenever it's good you can use it, and whenever it's bad you
can interpolate using the IMU. You can correct the current position
estimate with old visual data, essentially by first correcting the old
position estimate then doing an updated interpolation to current.

If you're using this to reconstruct the vehicle path after the fact, you
can interpolate in two directions, from both past and 'future' position

If the visual system just gives you deltas, and occasionally cuts out,
then you'll have to depend on tilt from the IMU combined with odometry
to estimate altitude change. The problem with any sort of scheme where
you measure deltas and get positions is that you'll be integrating
noise, so your position error will be ever growing.

Tim Wescott
Control system and signal processing consulting
From: Michael Plante on
Rich Webb wrote:
>It's not that you need GPS as such but you will need a sensor
>independent from the INS package to be able to estimate, e.g., sensor
>bias drift. The 3D visual system should work fine for that, to enable
>you do run an occasional update cycle where you compare estimated to
>observed position/orientation and make the necessary adjustments to your
>operating parameters.


One way you might be able to get a guess of how well this idea will (or
won't) work is if you, for the moment, ignore the visuals and just do
zero-velocity updates to see how tightly you'll eventually need to bound
your biases. This implies you can reliably know when you're stopped, and
preferably stop whenever you want, which may be difficult for the terrain
you hint at. If it works, then you can look at the more complicated
problem of using the cameras to update things w/o stopping.

Unrelated to the quoted text, I would also point out that you may be able
to make accurate turns with the IMU, since that's fairly short-term. The
limiting factor there may be a matter of knowing which way "down" is in
your body frame. This is not simply a matter of looking at the
accelerometers (though that's a decent first stab for some platforms).

How much development time you have might well govern further suggestions.
There are some fancy things people are working on that might work okay in
lieu of expensive sensors, but you'll be messing with them for awhile
(SLAM, etc.). This might address the absolute positioning problem if you
return to the same location occasionally, even w/o much prior knowledge of
your environment.

From: Jeff Cunningham on
On 3/9/10 3:11 PM, Alfred Bovin wrote:
> As I said, I have no previous experience with inertial navigation, but what
> is the typical usage of such inexpensive devices as the one I have? Is it to
> have short term navigation at a higher frequency than more reliable sensor
> data is available (such as GPS)?

No, it is to deploy the air bag when you crash your car.