What is your go-to machine for powerful/economic Ubuntu 240.4, ROS2 robotic development machine?

[James H Phelan]

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?):

Let's compare the Mac Mini with an M4 chip and a system with an Intel Core i9 processor, focusing on their suitability for running Ubuntu 24.04 and ROS2 robotics:

1. Hardware Architecture:

• Mac Mini M4: Uses Apple's custom-designed ARM-based M4 processor. This means it uses a different instruction set compared to the x86 architecture of Intel processors.
• Intel Core i9: Uses the traditional x86-64 architecture. 

2. Ubuntu 24.04 and Asahi Linux:

• Mac Mini M4: While Ubuntu doesn't officially support Apple Silicon Macs, projects like Asahi Linux and Ubuntu Asahi are actively working on enabling Linux distributions on these machines. However, full hardware support (including crucial components for optimal performance) may still be under development and not fully stable for the latest M4 chip. You can try installing Ubuntu via a virtual machine (like Parallels) if a native install proves difficult.
• Intel Core i9: Intel processors have excellent and long-standing support for Ubuntu and other Linux distributions, including the latest Ubuntu 24.04. 

3. ROS2 Robotics:

• Mac Mini M4: ROS2 can be installed on macOS using Homebrew. If you choose to run Ubuntu on a Mac Mini M4, either natively via Asahi Linux or in a VM, you would install ROS2 as you would on any other Ubuntu system. However, some users have reported better performance in VMs for ROS2, especially when needing to connect to external robots.
• 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. 

4. Performance:

• Mac Mini M4: The M4 chip excels in performance per watt, meaning it delivers great performance for its power consumption. In certain benchmarks, it has shown impressive results, potentially even surpassing Intel Core i9 in some single-core tests. However, its raw performance in some demanding tasks might be less than a top-tier Intel i9.
• Intel Core i9: Intel Core i9 processors are known for their high raw performance, especially in demanding, multi-threaded tasks. They are also well-established and widely supported in the Linux ecosystem. 

5. Other Factors:

• Expandability: Intel-based PCs generally offer more flexibility in terms of hardware upgrades compared to Mac Minis, where components like RAM and storage are typically not user-upgradeable.
• Cost: While an Intel i9 system might require additional expertise for setup and optimization for real-time performance, it can potentially offer a better performance-to-cost ratio compared to a Mac Mini.
• Linux support: Asahi Linux is making great strides, but full and stable hardware support for the M4 chip on Linux may still take time. 

Conclusion:

• If your primary focus is native Ubuntu 24.04 and ROS2 development, an Intel Core i9 system is likely the better choice due to its mature Linux support and established compatibility with robotic development environments.
• If you value performance per watt, quiet operation, and are comfortable with the ongoing development of Asahi Linux, the Mac Mini M4 could be a viable option, especially if you plan to use a VM for your ROS2 work. However, be aware of the potential challenges in terms of native Linux support and hardware compatibility. 

Ultimately, the best choice depends on your specific needs, technical comfort level with setting up and troubleshooting Linux on non-traditional hardware, and willingness to work within the limitations of the current support for Ubuntu on Apple Silicon.

RoverDoc

-- 
James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

[camp .]

> I asked Google AI and it said (Do y'all agree with this?)

    For RViz2 and slam_toolbox, or Nav2 (Humble), an Intel Core i7-4500U @ 1.8GHz is adequate. However, when attempting to add Gazebo simulation as well (Linorobot2), it is not adequate. I use an i9 Dell (6 cores, 12 threads) laptop (3.5GHz) for Linorobot, and it works fine (Gazebo, Nav2 or slam_toolbox, and RViz).

    In general, it's the same rule as always: as fast a processor and as much memory as you can afford.  :-]  

Enjoy,

Camp

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/7bd0c6d7-92cb-4ec7-a860-e808e941d387%40hal-pc.org.

[Stephen Williams]

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 650-450-8649 | fax:703-995-0407 | s...@lg.net | https://VolksDroid.org | https://BlueScholar.org | https://sdw.st/in

[Marco Walther]

On 6/14/25 15:43, 'Stephen Williams' via HomeBrew Robotics Club wrote:
>> *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.

That's why we finally have Pi5's with fans;-) In the near future, your
CPU might not need a fan but your GPU/NPU/AI-PU might;-)

>
>
> 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

It all depends, as always.

>
>
> 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.

Overall, I'm not a big fan of Apple devices. I opened a couple and
wasn't very impressed.

>
> The Jetson line is specifically tuned ARM-based system architecture to
> support things like robotics.

As long as you stay in the Nvidia garden. If you need an Nvidia Jetson
for something specific and you can live with it, there is nothing wrong.

> 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.

>
> I'll use big PCs to do ML training.  I am only interested in a high-end

For really large ML training, the choice is probably a cloud spot setup.
For smaller things, a local PC might be enough. But it really depends on
how much money & time you can rationalize;-)

I do have an Nvidia ADA in my laptop;-)

> 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.

I'm still living the Intel+Nvidia live for laptops;-) I did not need a
real desktop computer for 15++ years;-)

My current main laptop is an i9-13980HX with 64GB RAM and an Nvidia ADA
2000. I usually run a couple VirtualBoxes and some Docker containers on it.

But I also have a couple of Pi's ;-) From the P1B ... P5;-) Six++ of
them running 24*365;-) And some Odroid's & Jetson's ...;-)

-- Marco

>
>
> Stephen
>
>
> On 6/14/25 2:16 PM, camp . wrote:
>> > I asked Google AI and it said (Do y'all agree with this?)
>>
>>     For RViz2 and slam_toolbox, or Nav2 (Humble), an Intel Core
>> i7-4500U @ 1.8GHz is adequate. However, when attempting to add Gazebo
>> simulation as well (Linorobot2), it is not adequate. I use an i9 Dell
>> (6 cores, 12 threads) laptop (3.5GHz) for Linorobot, and it works fine
>> (Gazebo, Nav2 or slam_toolbox, and RViz).
>>
>>     In general, it's the same rule as always: as fast a processor and
>> as much memory as you can afford.  :-]
>>
>> Enjoy,
>> Camp
>>
>> On Saturday, June 14, 2025 at 04:22:09 PM EDT, 'James H Phelan' via
>> HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:
>>
>>
>> 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?):
>>
>> Let's compare the Mac Mini with an M4 chip and a system with an Intel

>> Core i9 processor, focusing on their suitability for running*Ubuntu
>> 24.04 and ROS2 robotics*:
>> *1. Hardware Architecture:*
>>
>> * *Mac Mini M4:*Uses Apple's custom-designed ARM-based M4 processor.

>> This means it uses a different instruction set compared to the x86
>> architecture of Intel processors.

>> * *Intel Core i9:*Uses the traditional x86-64 architecture.
>>
>> *2. Ubuntu 24.04 and Asahi Linux:*
>>
>> * *Mac Mini M4:*While Ubuntu doesn't officially support Apple

>> Silicon Macs, projects like Asahi Linux and Ubuntu Asahi are
>> actively working on enabling Linux distributions on these
>> machines. However, full hardware support (including crucial
>> components for optimal performance) may still be under development
>> and not fully stable for the latest M4 chip. You can try
>> installing Ubuntu via a virtual machine (like Parallels) if a
>> native install proves difficult.

>> * *Intel Core i9:*Intel processors have excellent and long-standing

>> support for Ubuntu and other Linux distributions, including the
>> latest Ubuntu 24.04.
>>

>> *3. ROS2 Robotics:*
>>
>> * *Mac Mini M4:*ROS2 can be installed on macOS using Homebrew. If

>> you choose to run Ubuntu on a Mac Mini M4, either natively via
>> Asahi Linux or in a VM, you would install ROS2 as you would on any
>> other Ubuntu system. However, some users have reported better
>> performance in VMs for ROS2, especially when needing to connect to
>> external robots.

>> * *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.
>>

>> *4. Performance:*
>>
>> * *Mac Mini M4:*The M4 chip excels in performance per watt, meaning

>> it delivers great performance for its power consumption. In
>> certain benchmarks, it has shown impressive results, potentially
>> even surpassing Intel Core i9 in some single-core tests. However,
>> its raw performance in some demanding tasks might be less than a
>> top-tier Intel i9.

>> * *Intel Core i9:*Intel Core i9 processors are known for their high

>> raw performance, especially in demanding, multi-threaded tasks.
>> They are also well-established and widely supported in the Linux
>> ecosystem.
>>

>> *5. Other Factors:*
>>
>> * *Expandability:*Intel-based PCs generally offer more flexibility

>> in terms of hardware upgrades compared to Mac Minis, where
>> components like RAM and storage are typically not user-upgradeable.

>> * *Cost:*While an Intel i9 system might require additional expertise

>> for setup and optimization for real-time performance, it can
>> potentially offer a better performance-to-cost ratio compared to a
>> Mac Mini.

>> * *Linux support:*Asahi Linux is making great strides, but full and

>> stable hardware support for the M4 chip on Linux may still take time.
>>

>> *Conclusion:*
>>
>> * If your primary focus is*native Ubuntu 24.04 and ROS2
>> development*, an*Intel Core i9 system is likely the better
>> choice*due to its mature Linux support and established

>> compatibility with robotic development environments.

>> * If you value*performance per watt, quiet operation, and are
>> comfortable with the ongoing development of Asahi Linux*, the*Mac
>> Mini M4 could be a viable option*, especially if you plan to use a

>> VM for your ROS2 work. However, be aware of the potential
>> challenges in terms of native Linux support and hardware
>> compatibility.
>>
>> Ultimately, the best choice depends on your specific needs, technical
>> comfort level with setting up and troubleshooting Linux on non-
>> traditional hardware, and willingness to work within the limitations
>> of the current support for Ubuntu on Apple Silicon.
>>
>> RoverDoc
>>
>> --
>> James H Phelan
>> "Nihil est sine ratione cur potius sit quam non sit"
>> Leibniz
>> --
>> You received this message because you are subscribed to the Google
>> Groups "HomeBrew Robotics Club" group.
>> To unsubscribe from this group and stop receiving emails from it, send
>> an email to hbrobotics+...@googlegroups.com.
>> To view this discussion visit https://groups.google.com/d/msgid/

>> hbrobotics/7bd0c6d7-92cb-4ec7-a860-e808e941d387%40hal-pc.org <https://
>> groups.google.com/d/msgid/hbrobotics/7bd0c6d7-92cb-4ec7-a860-
>> e808e941d387%40hal-pc.org?utm_medium=email&utm_source=footer>.

>> --
>> You received this message because you are subscribed to the Google
>> Groups "HomeBrew Robotics Club" group.
>> To unsubscribe from this group and stop receiving emails from it, send
>> an email to hbrobotics+...@googlegroups.com.
>> To view this discussion visit https://groups.google.com/d/msgid/

>> hbrobotics/1363318835.572936.1749935768906%40mail.yahoo.com <https://
>> groups.google.com/d/msgid/
>> hbrobotics/1363318835.572936.1749935768906%40mail.yahoo.com?
>> utm_medium=email&utm_source=footer>.
> --
>
>
> *Stephen D. Williams*

> Founder: VolksDroid, Blue Scholar Foundation

> 650-450-8649 <tel:650-450-8649> | fax:703-995-0407 <fax:> | s...@lg.net
> <mailto:s...@lig.net> | https://VolksDroid.org <https://VolksDroid.org> |
> https://BlueScholar.org <https://BlueScholar.org> | https://sdw.st/in

>
> --
> You received this message because you are subscribed to the Google
> Groups "HomeBrew Robotics Club" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to hbrobotics+...@googlegroups.com

> <mailto:hbrobotics+...@googlegroups.com>.

> To view this discussion visit https://groups.google.com/d/msgid/

> hbrobotics/7f9c06eb-1a36-4d5e-ae5d-f58ea48a98cb%40lig.net <https://
> groups.google.com/d/msgid/hbrobotics/7f9c06eb-1a36-4d5e-ae5d-
> f58ea48a98cb%40lig.net?utm_medium=email&utm_source=footer>.

[A J]

For everyday use I using an older i9 with an RMA return GPU. It seems okay for localized LLM and ROS

use is light right now. Last week I noticed the 265K listed at 300 and today for 259. It seems like a good

time to upgrade with 32 bit performance of gaming CPU about 3 - 4x of mine. I did update to 64 GB this

year but for Bot simulation in 2nd half a better gaming system would be preferred.

https://www.pcmag.com/news/intel-panther-lake-silicon-demo-computex-2025

[Sergei Grichine]

James,

I can only speak of my own experience, so here it is:

My main computer is a DIY - a 2011 "game quality" ASUS motherboard with Intel Core I7-2600K @ 3.4 GHz from the same year. I put an NVIDIA 3600 card in it. 16GB RAM.

My budget test box is an eBay Dell 9020 (~2013) - an Intel Core i5-4590 CPU @ 3.3GHz with Intel integrated graphics and 16GB RAM.

Both run Ubuntu 24.04, frequently updated. ROS Jazzy with all thrills - https://github.com/slgrobotics/robots_bringup/tree/main/Docs/ROS-Jazzy

As you can see, neither of the machines is a screamer, but I don't see a reason to upgrade.

The I7 does well for all tasks, but, strangely, refuses to run Nav2 containers (I run those on RPi 5 across WiFi then). Strange installation problem, I guess.

The I5 struggles with Gazebo Sim, but can still be used for testing. It runs everything, but is slow.

What is important IMNHO:

- motherboard reliability - anybody who remembers "capacitor plague" (https://en.wikipedia.org/wiki/Capacitor_plague) knows what I am talking about.

- Having many virtual cores (I7 = 8) is very important for ROS2 with its reliance on Linux threading. An 8 virtual core machine (4 physical cores, like my I7) seems to be a minimum.

- A 3060 class video card is sufficient for Gazebo. I'd guess it is a minimum for comfort. BTW, laptop GPUs are not what they claim to be.

- 16 GB RAM is sufficient for all tasks. Don't forget to set up 16GB swap space.

- the quality of your network and WiFi is VERY important. My WiFi 6 tests at 700+ Mbits/s and some tasks (PointCloud from OAK-D Lite) would still saturate it. Test yours well - https://github.com/slgrobotics/robots_bringup/blob/main/Docs/Sensors/WiFi_Logger_Visualizer.md#wifi-testing-and-benchmarking

- having two machines makes troubleshooting easier. Especially for "strange" problems.

Two points to keep in mind:

1. Some recent Intel I9 CPUs (13th and 14th gen) are notoriously prone to failing - you can find which ones. Thermal, voltage or whatever - they are just too much of a problem out of the box.

2. There's a flood of Windows 11 non-compliant machines on eBay, which would run Ubuntu perfectly well. I'd suggest looking at those if you want a secondary ROS box. Video cards aren't cheap still. New "security features" (TPM) could be a blessing or curse for Linux machines in the long run.

To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/804c4f12-3460-4d55-888c-af9db6fd09f3n%40googlegroups.com.

--

Best Regards,

-- Sergei

[Michael Wimble]

Here is what I use:

I build both my robot SIgyn and my Linux desktop from 

AMD Ryzen 9 7900X Desktop Processor without Cooler 12 Cores 24 Threads 4.7GHz Base Clock 5.6GHz Max Turbo 170W TDP AMD Radeon SKU: CPUAMD7900XR 1 $369.99 Corsair CMK64GX5M2B5200C40 Vengeance 64GB DDR5 2x 32GB Memory Kit 5200MHz C40 40-40-40-77 1.25V XMP 3.0 Black SKU: MEMCOR64524R 1 $139.99 ASUS ROG STRIX B650E-I GAMING WIFI Mini-ITX Motherboard Supports Ryzen 7000 Series Socket AM5 LGA 1718 Max 64GB DDR5 SKU: MOTASU650EIR 1 $299.99 Subtotal  $809.97 Shipping & Handling  $0.00 Tax  $73.91 E-Waste  $0.00 Order Total $883.88

My desktop has an old NVidia graphics card but I run rviz2 and gazebo on both machines and it is barely loading them. So, yes, gazebo without a graphics card. This AMD processor has built-in video. I built both the same in case I ever burn out something or need spares. I can canabalize something while I wait for spares. I bought the systems at Central Computers in the Bay Area—they have good prices, are knowledgeable and have been trusted partners for me. I use the MiniItx format because the roughly 7” square format fits nicely in my robot. I have 12v, 5v and 3.3v brick converters in my robot that convert the 42v from my battery down to the 3 supply voltages needed by the motherboard. I used to just bring out the wifi with the provided antennas that come with the motherboard, but, like Sergei, I’ve decided to use an external wifi box, mostly because I  need a specific reconnect behavior to the wifi mesh nodes around my house and the built-in chips don’t seem to do this, and I don’t particularly want to have to write code to force the behavior.

It’s fast, reliable, and within my budget. I like to try to buy things once, as much as possible. I’ve spent thousands of dollars over the last few years trying various CPU combinations and I’m pretty darn pleased with this setup. Add in my two custom Teensy 4.1 boards and a PI5 with the AI accelerator, and it all works a treat.

BTW, I’m designing yet another version of my Teensy 4.1 board which I may offer to the club if they are interested. There’s a lot of IO pins, a lot of memory, and a fast clock. The Teensy 4.1 is my goto for anything requiring predictable IO control.

[Michael Wimble]

Oh, I should also point that that a huge consideration for me was power use by the computer. This configuration has about a 60 watt maximum power use. With my 42v, 30Ah battery, I can run a long time. This is on order of the power that my motors use as well.

[Stephen Williams]

On 6/14/25 6:19 PM, Marco Walther wrote:

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:

https://termux.dev/en/

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

[Chris Albertson]

Just run the Linux system in a Virtual Machine.

Then you can host the VM on a Mac mini.  But the VM is portable, and later if you get some other computer, you can move the VM.  Being portable is kind of a big deal, you can share the VM with other people. Or transparently upgrade the hardware.    Backup is easy, too as a VM is just a file.

That said the most powerful computer you can get will be a Xeon workstation that just came off-lease.   Spend about $800 on one then install one or more Nvidia GPUs.

But in the end, if compute power is what you need ROS works on a network just fine, so you can distribute the system over 3 or a dozen computers, whatever is needed.

I’d say to set up a VM on whatever computer you have, even a Windows PC notebook, and then put Ubuntu on the VM and then ROS2.    Use that until you see a reason to upgrade.

[Sergei Grichine]

I just updated my Intel I5-4590 test box and ran the Turtlebot demo: https://github.com/slgrobotics/robots_bringup/blob/main/Docs/ROS-Jazzy/README.md#verify-your-installation

It took about a minute to start, but everything worked fine, and the bot navigated as it should. No crashes or slowing down at all. Yes, a bit busy oh the CPU side, but only 3/6 GB memory used.

So, ROS2 runs well on an average 12 years old machine. But the development machine, of course, should be much beefier. And don't forget to treat yourself with a 4K screen https://www.amazon.com/dp/B08TZVQBT5 ;-)

[Chris Albertson]

What people are missing here is the intended use of the computer. OK, you want to run Ubuntu and ROS2. But is this s development system or does it go inside a robot, and then what kind of robot and how big is it?

You don’t want a Raspberry Pi as your desktop computer, they are dead slow for general usage, but also you don’t want to put a Mac Mini inside a small robot. You want to find a match, none is ‘best”

I have settled on running Ubuntu inside a VM on my Mac for most development that needs Ubuntu. But you can do a lot or development using macOS with no need for Ubuntu or ROS

macOS, with no virtual machine, is a surprisingly good system for AI and machine learning because it uses “unified RAM” that is shared by the CPU and GPU but if all-out performance is needed then you must pay for some high-end Nvidia GPS but you call have to pay quite a lot and a Google or Amazon cloud server might be better.

[Stephen Williams]

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

650-450-8649 | fax:703-995-0407 | s...@lg.net | https://VolksDroid.org | https://BlueScholar.org | https://sdw.st/in