Team,
I've heard several of y'all talk about their favorite economical yet powerful computer.
I've heard about i9 COREs and Mac Mini M4.
A Mac Mini M4 runs $500. An Intel i9 processor alone, without the box, costs that.
A full machine like the Dell Envy
run $1600.
I asked Google AI and it said (Do y'all agree with this?)
I'm concerned about Linux on the Mac and don't do Dockers (yet?):
RoverDoc
-- James H Phelan "Nihil est sine ratione cur potius sit quam non sit" Leibniz
Intel Core i9: Ubuntu with x86 architecture is generally considered the preferred platform for ROS2 robotics development, especially when working with widely used development kits like NVIDIA's Jetson AGX series, which are ARM-based. Using an x86 system with a VM or container (like Docker) effectively provides the same development environment as the robot's runtime environment, while taking advantage of the Intel processor's performance.
This paragraph is a big garbled. The x86 is preferred especially when working with... an ARM-based system? Emulating an ARM on x86 is taking advantage of the Intel processor?
Most things are going to some form of ARM, especially anything mobile, LoSWAP (low size, weight, and power) + cost, compact, low heat, etc. I would assume any computer actually on a robot to be ARM based. Intel is useful for large, heavy desktops. It is OK in a laptop, but not competitive with ARM-based systems if you care about battery life, avoiding fan noise, overheating, etc.
For robotics, we're usually interested in SWaP, usually Low-SWaP aka LoSWaP. And also usually cost, so LoSWaP-C. This is a random page on it, which happens to be from a company with some nice positioning units:
https://www.pnisensor.com/swap-c-size-weight-power-and-cost/
https://www.baesystems.com/en-us/definition/what-is-swap-c
On MacOS, Parallels has some very specific integration with MacOS
that supports very efficient I/O in VMs. There are a number of
ways to run VMs, but last I looked into it, Parallels had a big
edge. If you are not emulating a processor, a VM shouldn't have
much overhead.
The Mac Mini M4 keeps wowing reviewers with performance given price, size, and power needed.
The Jetson line is specifically tuned ARM-based system
architecture to support things like robotics.
But Qualcomm chipsets are what are in most mobile phones. They
are very good, as would be a lot of mobile phones. But both with
ML and integration, the Jetson may often be a better choice. On
the other hand, using slightly older phones is probably the best
price / performance vs power + peripherals option.
I'll use big PCs to do ML training. I am only interested in a
high-end MacBook Pro for a laptop. For deploying anything
serious, probably a Jetson if I have the budget or a phone or
another board similar board if I'm holding price down. There are
many choices for auxiliary processors. It is time to take another
look at that, think about required interconnects etc. Things like
support for 2 wire automotive Ethernet at high speeds, USBC, will
likely sway me.
Stephen
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/1363318835.572936.1749935768906%40mail.yahoo.com.
|
Stephen D.
Williams
Founder: VolksDroid, Blue Scholar Foundation |
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/804c4f12-3460-4d55-888c-af9db6fd09f3n%40googlegroups.com.
On 6/14/25 15:43, 'Stephen Williams' via HomeBrew Robotics Club wrote:
But Qualcomm chipsets are what are in most mobile phones. They are very good, as would be a lot of mobile phones. But both with ML and integration, the Jetson may often be a better choice. On the other hand, using slightly older phones is probably the best price / performance vs power + peripherals option.
If you have an old phone, it might be a choice for a robot. But it's probably hard to get Linux & ROS2 onto it. I used Odroids for some time, they took the phone CPUs and put them into SBC with support for Linux & Android. That worked ok in 2015.
Android is already Linux, just a bit hacked up. You don't want
to try to install Linux directly on most mobile devices. It is
possible, but it may be difficult without the OEM package, a
Qualcomm OEM dev board that matches, etc. Just seldom worth it.
I worked with that level at one company, modified & built our
own version of Android with additions to the camera API, etc.
There is little that you can't do with Android, especially if you
root the device - but that isn't usually needed anymore. Just
side load a Java / Kotlin / NDK C/C++ app.
Android NDK C++ multithreaded apps can use most Android resources. You can have additional programs that can be started. This is probably the best proof of that, and potentially useful:
Seems like ROS2 is close to running on Android:
https://www.danaukes.com/notebook/deprecated/build-ros-on-android/
This shows what sensors are available at the Linux-within-Android
level:
https://wiki.termux.com/wiki/Termux-sensor
Also related:
http://download.ros.org/downloads/roscon/2022/Building%20ROS%202%20enabled%20Android%20apps%20with%20C++.pdf
This looks potentially interesting. I think it is just the viewer connecting over ROS1/2 bridge.
https://gitee.com/ncnynl/ROS_Flutter_Gui_App
Probably makes sense to run the main ROS2 node on a RPi5 or
something, with Android publishing / consuming events. I see that
some people just wrote DDS code directly. And other approaches.
However, on a very price-sensitive project, should be able to use
just an Android phone for nearly everything.
For automotive Android, there is work on full support:
https://spyro-soft.com/developers/android-automotive-vhal-with-ros-2-case-study
sdw
-- Marco
Stephen

Looks like a nice monitor. 4K is a necessity for me whenever
possible. I switched to using cheap but high quality 4K TVs a
while ago. Put them in PC / game mode to remove latency of image
postprocessing they default to. They usually are 43"+, so you
need some room. At my main desk, I have a 55" 4K TV leaned up
against the wall, plus laptops and other computers. That works
better than you would think.
The Amazon FireTV series, made by at least a couple
manufacturers, seems to have a pretty good minimum quality with
very low margins. Beware there are a number of "4K" TVs that are
faking it somehow, not really at 4K. Fry's had a few of those in
stock that were terrible.
The FireTV Omni is an upscale version of that line:
https://www.amazon.com/introducing-amazon-fire-tv-43-inch-omni-qled-series-smart-tv/dp/B09N6LRBTR
4-series, seems to be newer version of the standard model:
https://www.amazon.com/amazon-fire-tv-43-inch-4-series-4k-smart-tv/dp/B0CZ9WV2ZX
Some versions are very inexpensive now: $169.99
https://www.amazon.com/insignia-fire-tv-43-inch-class-f30-series-4k-smart-tv/dp/B0CMDJ8TK3/
I did buy a smaller 4K professional monitor, but will have to
track down model & source.
sdw
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/CA%2BKVXVMr-eRcMdkMmjOCGDKDSA-12S6PjkTtxB0eu5g9cY3bsA%40mail.gmail.com.
|
Stephen D.
Williams |
Founder: VolksDroid, Blue Scholar Foundation |