Hi Eric,
Thanks for the analysis. Although I was disappointed to hear that my
robot wasn't *really* traveling at the speed of light. :) And thanks
for the heads up regarding the dt calculations--I *thought* I was
handling missed data correctly but now I'll double-check. In the
meantime, I'll look at the bag files myself so I can see if I see what
you see--seems like a good learning opportunity to see when something is
going wrong.
Thanks again to you and Eitan.
--Patrick
> Patrick,
>
> I took a look at your bag files and have discovered a pretty serious
> looking
> problem with your odometry at 20Hz. When plotting your reported angular
> velocity (rxplot /odom/twist/twist/angular/z), I noticed many large
> spikes
> in it (faster than -25 at one point... that's a _crazy_ number of
> radians/sec :) ); this type of noise is definitely not a good thing and
> might be throwing off the navigation stack (pretty sure it assumes you
> have
> well-behaved odometry data). Since that magnitude of spike isn't
> present in
> the 10Hz version, you may want to make sure that your velocity
> calculations
> are not making assumptions about loop closure rates for their dt. For
> example, if every once in a while, due to timing problems you actually
> see
> 2-3 dt worth of angle change in a single loop and then divide by dt, you
> will get a completely incorrect velocity; I saw behavior just like
> that on
> my real-time controller when we accidentally stuck some non-deterministic
> code in the real-time loop.
>
> Also, the -o flag for rosbag record can be handy when sharing bag
> files - it
> can give a descriptive prefix to the bag file, such as
> 'odometry_at_20_hz...'. See
> http://www.ros.org/wiki/rosbag/Commandline#record for details.
>
> Hope that helps.
>
> - Eric
(text/plain)