[Dprglist] Visualizing data [was: PID]

[Murray Altheim]

On 4/10/20 11:29 am, David Anderson via DPRGlist wrote:
> Hey Murray,
> 
> Sounds like great progress.

Hi David,

Well, it doesn't feel like it quite yet, but as you know sometimes it takes
a bit of tooling before one actually gets down to the real task. So I'm
looking at the likes of GnuPlot or Grafana as tools to better see what's
happening. Staring at console outputs only goes so far.

> I also log robot data on board in real-time and then dump it at the end 
> of the run.  The command on the PC end that fetches the data also makes
> a gnuplot "*.plot" file and invokes gnuplot with that as an argument. 
> So I type in "get_log" and a graph pops up.

If I can manage to get the Python library to work, at the end of the run it
will actually do an http POST and upload the CSV directly to Elasticsearch,
where if I choose the same ID value for the data will deliberately overwrite
the existing record. Then if I've got Grafana pointing at that record I can
automatically refresh the visualisation. So if I manage to get that all
running it will be almost entirely hands-free, apart from hitting F5 to
refresh the screen (which might not even be necessary, dunno, I think Grafana
can probably do live updates).

> The graph that Bob Cook sent is exactly what we were looking for. (thanks 
> Bob!).   A couple of comments. Note that his PID is responding to a large
> step function, no slewing at all of the input.  This is the test that you
> want to run.   
> Disable any input slewing.  It just makes the PID tuning harder.

Yyes, thanks, understood. In my messages here I might be confusing things,
as my end goal is this "one meter behaviour" where I want to tune the whole
behaviour. This is distinct from the PID tuning, which is a prerequisite
for the behaviour. It's just that the first time I ran "One Meter" the robot
came to rest and then did its fidgety hunting back and forth thing, which
was when I noticed this.

> It's useful to reflect that the PID controller must be able to react
> quickly to step changes in the environment, even if the control input
> itself is correlated and smooth.  So running on a rough surface, or 
> when a wheel drops into a depression for example, the feed back velocity > coming from the motor encoders may change a lot, and change rapidly.
> That means the proportional error will also change a lot and change 
> rapidly, even if the control input is smooth.  The PID controller has
> to be able to react quickly to counter those changes.

Understood. And tuning my PID controller on the bench with the wheels
unloaded is I'm guessing not going to quite do it, that the robot needs
to be sitting on the ground and actually moving. I've been doing my
preliminaries on the bench then putting it on the wood floor to fine
tune.

> So acceleration and deceleration ramps that present a smoothly changing
> input to the controller can't be implemented by making the PID controller
> itself sluggish overall. Or rather, they can, but it's probably not  > what you want  :)

Understood -- I'm not doing any PID tuning while trying to solve that
One Meter behaviour.

> Every robot builder should have that Flynn/Jones book on their shelf.
> A classic.  If you haven't already seen it there is another, later book
> by Joseph Jones, "Robot Programming: A Practical Guide to Behavior 
> Based Robotics" which is also excellent, a bit more technical and 
> focused on subsumption.

Thanks, I've seen that on Amazon but didn't connect it was the same Jones
as the other book.

I've also collected some references to Rodney Brooks' papers related to
subsumption and BBS:

   https://service.robots.org.nz/wiki/Wiki.jsp?page=RodneyBrooks
   https://service.robots.org.nz/wiki/Wiki.jsp?page=References

Maja Mataric was one of Brooks' students at MIT and pioneered what is
called Socially Assistive Robotics (a sub-field of human-robot interaction)
and has a few books to her name, "The Robotics Primer (Intelligent
Robotics and Autonomous Agents series)", which I haven't read but looks
interesting. It is available as a PDF at:

    http://www.ict.griffith.edu.au/~vlad/teaching/robotics.d/RESOURCES/mataric-primer.pdf

She is a co-author of "Embodiment in Socially Interactive Robots (Foundations
and Trends(r) in Robotics)". That's an $80.00 academic paperback but also
available again as a PDF:

    https://viterbik12.usc.edu/wp-content/uploads/2019/06/Mataric%CC%81_Embodiment-in-Socially-Interactive-Robots.pdf

Finally, there's "Probabilistic Robotics" by Sebastian Thrun, also available as
an expensive academic hardcover ($65) or as a PDF:

    https://docs.ufpr.br/~danielsantos/ProbabilisticRobotics.pdf

I think I'll copy these over to my Kobo Reader and tackle them on the train
to work this week... thanks for tickling me to go out and look for these.

There's been quite a lot of research in the "low end", affordable side of
robotics (compared to what's done by NASA, industry, etc.) and a lot of my
interest is in seeing (eventually) what I can do with a relatively cheap
($500-$1000) robot and access to all this research. It feels like an area
that still has lots of room for exploration...

Cheers,

Murray

...........................................................................
Murray Altheim <murray18 at altheim dot com>                       = =  ===
http://www.altheim.com/murray/                                     ===  ===
                                                                    = =  ===
     In the evening
     The rice leaves in the garden
     Rustle in the autumn wind
     That blows through my reed hut.
            -- Minamoto no Tsunenobu