On Tue, Jun 30, 2015 at 1:39 PM, Robert Schwebel
<r.schwebel@...> wrote:

On Tue, Jun 30, 2015 at 01:30:04PM -0300, Gustavo Sverzut Barbieri wrote:

On Fri, Jun 26, 2015 at 11:47:45AM -0300, Gustavo Sverzut Barbieri wrote:

Not yet. We're using an internal one, github issues for the public. Is
there a good free ML service you recommend?

I don't think so. Let's stay here for the moment :-)

Our first impression when we saw soletta was: wow, they did it right.

That's nice to read. You mean just the build or the project itself? Share your
ideas and needs please :-)

The project itself. We are doing a lot of industrial automation
projects, and over the years, everyone invents the typical IO framework
stuff over and over again. Even worse, people with a different
background invent the same stuff in different ways:

- control people use Matlab/Simulink
- measurement guys use Labview
- factory automation does IEC61131 or IEC41499
- IoT folks use COAP etc.

Everyone reinvents wheels all over the place already, and now the IoT
people start working on the same things again.

There are many frameworks out there which have started with a university
background and have many deficiencies for professional use, and on the
other hand there are commercial frameworks which take things like safety
into account, but are not open source.

It's all a big mess.

yes, this is the case. And to make things worse, some assumptions from
Industrial or "old" embedded are making hard to get efficient user
consumer electronics. Take polling, for instance. Most software out
there is based on a fixed base tick, like Arduino or even some OS. If
you can't define timers, you can't reprogram your base tick and thus
cannot do a technique similar to Linux 'tickless' scheduler. That
means battery drain. Not an issue for industrial where you have
reliable power supply with backups/ups and where the lack of power
stops everything else. On homes, power outages are common, people want
to run on few duracell AAA batteries and so on... Network connectivity
is also similar, particularly if you take the wifi mess into account.
And upgrades, while before either you'd never upgrade or you'd do so
using experienced maintenance support, while with consumer electronics
you get get your inexperienced family member to do it and they have
nobody to call if things go wrong.

That's why we also plan to cover these bits with Soletta. We will
implement some sol_platform targets to reset to factory defaults,
emergency and upgrade for platforms we support, of course some rely on
having a hardware (i.e.: if we do have flash we can upgrade, if we
have EEPROM/NVRAM/... we can reset to factory defaults). Those would
be generic implementation, users can always provide their own and
still use the common API.

Did you take some time to try the linux-micro platform with posix
mainloop? We're using it for some demos with impressive footprint
(140kb userspace using bluetooth low-energy - bluez-peripheral +
muslc) And we can avoid using busybox or shell scripts :-) a single
binary does it all, including some services such as mounting from
/etc/fstab or setup ipv6.

IMHO what we need is a heterogeic hierarchical modeling framework, in
the spirit of Ptolemy II (but withaut Java), coupled with the practical
experience from GStreamer and v4l2 subdevices, but for process data.

soletta comes close :-)

great, let's work to do it :-)


--
Gustavo Sverzut Barbieri
--------------------------------------
Mobile: +55 (19) 99225-2202
Contact: http://www.gustavobarbieri.com.br/contact