The Jetson Nano & Remote Desktop

I think I'm going to wait until I do more studying on this.   It started out as a $99 Jetson development "kit" and ends up being $199 package of various attachments to get it to work.   I too was thinking of using a power supply that I already had laying around.  But for $8.99 the official power supply for the board is the least expensive of the peripherals.

I don't plan on doing the Jetbot project, so I'll probably go with the 64GB SD cards.  I can always change over to 128GB later if I find that I need more space.   The extra space needed for the Jetbot software may have a lot to do with ROS. 

I'll probably go with the Intel 8265 WiFi card to just because the Nano comes with a built-in slot for it.

So the only thing left to decide on is the camera.   I would like to have a camera that is also compatible with the Raspberry Pi.

In the meantime I've also decided to add this acrylic case with cooling fan for another $15.   May as well have it all set up really nice after spending $200 for it.  

And I still don't have a keyboard or monitor for it.  I was hoping to do this all headless with Remote Desktop.   But after all my research on that I'm not sure that will even work.   If I have to buy a monitor for it that's going to be another bunch of money.

I was attracted to this in-part because it was only $99.   But it's really more like $200+ by the time you get it all set up to actually do something with it.   I hope it pays off in the end and can actually do the things I'm hoping to use it for.

I'm also wondering whether a Raspberry Pi 4 might work out better for my purposes?  It would be less expensive and also be guaranteed to be usable with Remote Desktop.   I think I'll go look into setting up a Raspberry Pi 4 in a similar fashion and see what that comes out to be as a final cost.

The  two main attractions to the Jetson Nano for me are the following:

1. More powerful GPU processor.
2. Lot's of A.I. example software that runs on it.

But in the end I don't even know if I'll really need the GPU power.  And many of the A.I. examples may potentially run on a Raspberry Pi 4.

It's really hard to make a final choice when you're as ignorant as I am about these products and exactly what their capabilities and limitations are.  I know that NVIDIA is heavy into the A.I. end of things, and that's where my interest is.  That's the other reason (probably the main reason) that I'm even looking at the Jetson Nano.   It appears to be designed with A.I. in mind.

I've decided to take this Jetson Nano A.I. course by Paul McWhorter before pressing on the checkout button on my Jetson Nano order.

I've only watched the first lesson so far.  What a MESS!  But he did get it all up and running.  By the way, I meant that the Jetson Nano is a mess, not Paul's lesson.  In any case he gets it all up and running and shows how to turn the annoying passwords off.  I don't need passwords for what I'm doing.  I'm just trying to learn about A.I.   It's not like I'll have my banking info on the Nano.  Not that I have any money anyway. ? 

But yeah, I like Paul's style so hopefully I'll learn a little bit about doing A.I. on the Nano and I'll be in a better position to know whether this is going to work for me.

He lists pretty much all the parts I already have in my shopping cart.  But he also includes a keyboard and mouse, but no monitor.  I guess he assumes that you already have a monitor which I don't.   I really want to set this thing up headless via Remote Desktop.   Not just to save on having to buy a monitor for it, but it's just more convenient for me all the way around.  Unfortunately his lessons don't go into that aspect of it.

None the less, he does promise to address the A.I. capabilities so I'm looking forward to learning about those features and functions.

I just noticed that Paul is promising to cover running the Jetson Nano "Headless" in his Lesson #4 which he hasn't recorded yet.    Although "Headless" doesn't necessarily mean Remote Desktop.   That could be just access to the command prompt.    So I'll have to wait and see whether he includes Remote Desktop in the lesson or not.

If I can see it working with Remote Desktop I'd feel much more confident about clicking on the Checkout button on my order.  It's THAT important to me. ? 

I guess I hit the jackpot finding Paul McWhorter's videos on the Jetson Nano and A.I.   Apparently he's just starting his lessons on this, but here's a really good preview of where he'll be going:

I still don't know about the Remote Desktop, but at least he's answered my questions about a compatible Python IDE for the Jetson Nano.    Apparently you need to use a particular IDE to be able to take advantage of all the unique features of the Jetson Nano.   So that's really encouraging because I certainly didn't want to write Python programs in a Linux terminal text editor! ? 

So it looks like Mr. McWhorter has done all the hard work for me and will be a great teacher.   So this worked out just in the nick of time for me.  He's just starting his lessons and I'm just ordering a Jetson Nano.

Also notice that he's already written his first Python program from scratch to track an object and get it's x,y coordinates.   So he's already where I'm hoping to go.   So I'll be glued to his channel for the A.I. lessons on the Jetson Nano.

I hope I can use Remote Desktop eventually, but if not, I suppose it will still be worth it if I have to settle for a dedicated keyboard, mouse and monitor.   The capabilities of the Jetson Nano for A.I. work are just too good to pass up. 

And yeah, the 128 core GPU is a large part of the attraction.   According to Paul we can take advantage of those 128 cores using the Python IDE he's using.  

So look out Jetson, here I come! ? 

This is exactly what I was  hoping for.  I want to write my own programs in Python from scratch.  I don't want to have to depend on large programs that other people already wrote.   That wouldn't be developing A.I.  That would just be using the A.I. someone else wrote.   And my whole goal  is to develop  my own A.I. systems.

So now I'm as excited as Paul.    Almost to the point of clicking on the checkout button!

But not quite.  I just want to make sure I get precisely what I want.  I don't want to order the wrong stuff cuz I'm a poet and pauper.  ?   No moola here!  I'm going out on a limb to invest $200 in this.  But it looks to be well worth the investment.   I really want to work with A.I., and the Jetson Nano appears to be the best product on the market for this, at least in this price range.

Posted by: @robo-pi

But it looks to be well worth the investment.   I really want to work with A.I., and the Jetson Nano appears to be the best product on the market for this, at least in this price range.

So all that time I spent learning to use the RPi has gone to waste!!

Continual learning without time to actually make use of it.

How does the Jetson compare with a little laptop for A.I. that I already own and can program?

Posted by: @casey

So all that time I spent learning to use the RPi has gone to waste!!

Absolutely not.  In fact, I actually intend on  using a RPi  as the main processor for my R.I. intelligence in my robot.   The RPi will simply access the Jetson Nano for information concerning things that require massive data crunching.   But I intend to use the RPi as the main "brain" of my robot.  You actually don't need a lot of speed for that. 

Posted by: @casey

Continual learning without time to actually make use of it.

I think that's where we're all at if we are attempting to develop something new and fully understand exactly how it works.    The other option is to simply use A.I. packages that have already been designed by others and employ their functionality without any need to know how they do what they do.

By the way, I'm not suggesting that one of these methods is necessarily better or worse than another.  They are simply two different things.   The former is what developers do, and the latter is what end-users do.  Both have their place.  If there were no end-users the developers would have no need to develop anything. ? 

My interest is in developing A.I. software and hardware.  So that's what I'm looking into.  I have a lot of ideas that I would like to explore that I haven't seen other developers focusing on.  So I need the tools that will allow me to develop those ideas.

Posted by: @casey

How does the Jetson compare with a little laptop for A.I. that I already own and can program?

The the single main feature that I see that sets the Jetson Nano apart from other computers, including laptops is the 128 core GPU.   It appears that this is what allows for this kind fast video processing to occur.  And it doesn't even need to necessarily be video processing.   You can process any high-level mathematics very quickly on this 128 core GPU.

The second feature is that there is a lot of software already written for the Jetson Nano that takes advantage of the 128 core GPU.    So even if you added a 128 core GPU processor to a notebook you'd be hard-pressed to find any software that runs on a notebook computer that would take advantage of this processing power.

Finally, and this is key for me.  As Paul points out in his Live Shop Talk 10 video above, he has already put together packages including a Python IDE that will allow the programmer to take full advantage of the 128 core GPU.   So now you can write your own programs to take advantage of this awesome processing power.

By the way, I'm not suggesting that anyone go down this path.  I'm just posting what I'm doing in case anyone else might be interested.  I'm also not suggesting that anyone should retire their Raspberry Pies.  I'll still be using my Raspberry Pies as the heart of my R.I. system.  I intend to have the main "Intelligence" running on the Pi with the Jetson Nano doing the "grunt work".

In fact, that's a whole other topic.   When I start making videos I'll be attempting to make a distinction between A.I. (artificial intelligence) versus R.I. (real intelligence).   In my proposed system the Jetson Nano will be processing the A.I. while the Raspberry Pi will be processing the R.I.

As an example, locating an object in a video graphics and tracking it is A.I. (artificial  intelligence). In other words, A.I. isn't really intelligence at all.   It's just any automated system that produces results that you want.  The program itself has no clue what its even doing.

On the Raspberry Pi I'll be running R.I. (real intelligence).  In other words, the software I'll write for the Raspberry Pi will actually know what it's doing.   Note: These are my own definitions that will be made technical.   What any software itself might actually "know" I'll leave up to the philosophers to argue about. ? 

But yeah, don't throw out your Raspberry Pies!  They have their place.   Or if you do throw them out throw them my way and I'll put them to good use. ? 

Today tried out Hello AI world on the jestson nano

Imagenet-canera googlenet is very intresting,  Tensor Flow fundamentals are very good.

Finally when i started the deep learning sequence i was so thrilled ! The camera recognized which breed my pet dog is! With 87% accuracy!

Posted by: @sdey76

Finally when i started the deep learning sequence i was so thrilled ! The camera recognized which breed my pet dog is! With 87% accuracy!

It is amazing what it can do.

I figure that if it can recognize dogs right down to a specific breed it should be able to recognize individual people too.   Especially if you only require that it be able to recognize a handful of specific individuals.  The only problem with the method they are using is that you need to have a thousand photos of your friends and family to train it with.   So that part can be a bit of a problem.

The program you are using was most likely trained on literally billions of images of dogs.     It also most likely wasn't trained using a Jetson Nano.  The Jetson Nano is just running the already-trained ANN. 

Posted by: @robo-pi

Posted by: @casey

Continual learning without time to actually make use of it.

I think that's where we're all at if we are attempting to develop something new and fully understand exactly how it works.    The other option is to simply use A.I. packages that have already been designed by others and employ their functionality without any need to know how they do what they do.

Well I am not a developer and I don't think Bill's DB1 is going to incorporate his own software inventions instead my impression is he will be using ROS to do the heavy lifting. Even when we develop something new we are using hardware or software without knowing how they work. We build new things out of already existing functional modules.

Loki is the kind of robot I would have liked to build.

This is Dave Shinsel's latest effort.
http://www.dshinsel.com/sheldon/

Regarding facial recognition there is a cognitive disorder, prosopagnosia, that is specific to facial recognition that shows how there are some things we fail to learn unless the machinery is there to begin with. With machines that recognize faces I thought they used biometrics to map facial features from a photograph or video after the face was itself recognized as a face. I don't think they train an ANN for every face to be recognized?

I was just talking with Paul McWhorter about his upcoming video on running the Jetson Nano headless.  He's only going to be covering simply SSH using Putty.  I already know that's possible.  He confirms that trying to use the Remote Desktop is plagued with OS problems.   He's going to wait until NVIDIA updates their OS image to include Remote Desktop capabilities.   Apparently they've had a lot of requests for this so they are supposedly working on it.   So I guess I'll just have to wait for that to come up.   In the meantime I'll need to look into getting a monitor for the Jetson Nano.  Dang!   Not it's going to be even more expensive! ? 

Posted by: @casey

Loki is the kind of robot I would have liked to build.

Yes, that is a very nice design for a physical body and head.   Loki is about the same size as I am hoping to build Alysha.   She too will have a wheeled base about the same side as Loki's.  And I would love to have two arms that are as capable as Loki's.   About the only real difference is that I'm planning on using a  human-like head pictured in m avatar as the face of Alysha.  But that project along is going to be a major undertaking.

Posted by: @casey

With machines that recognize faces I thought they used biometrics to map facial features from a photograph or video after the face was itself recognized as a face. I don't think they train an ANN for every face to be recognized?

That's the method I'm interested in working with.  But  you still need something like the Jetson Nano to provide you with the recognized face coordinate.  You then take that box and focus in just on that info to apply the biometrics to a small lesser portion of the captured picture.

There are also many different methods of mapping the biometrics.  That too can be done with or without ANNs.  Either way you still need to have a lot of biometric data of the subjects you are hoping to recognize.  So this doesn't free you from still needing a lot of pictures of the target face. 

Posted by: @casey

Well I am not a developer and I don't think Bill's DB1 is going to incorporate his own software inventions instead my impression is he will be using ROS to do the heavy lifting.

That's an interesting thought.  I don't know where Bill is headed with DB1.   I'm also not sure how much ROS will be a substitute for A.I.   I'm glued to the edge of my seat to see where Bill goes with DB1.  I'm sure it will be a very cool robot no matter how he proceeds.   I'm also sure that I will learn very much from following his lectures.   So even if he doesn't do precisely the same things I plan on doing his lectures will still be helping me quite a bit.  His lectures are extremely helpful when it comes to learning how to interface with various sensors, modules, and making inter-computer connections.

Also, if he goes the ROS route, then I'll learn all about that as well, as I'm sure Bill will explain every detail of it from the ground up as he always does with all his presentations.

So I won't be moving "over" to Paul McWhorter's channel.   I'll just be adding Paul's channel to my list of ever-growing resources. ? 

Posted by: @robo-pi

The extra space needed for the Jetbot software may have a lot to do with ROS. 

Precisely

That's why my other 2 cards can be only 64g, while the Jetbot is 128. It's the ROS that's taking up most of the space

Posted by: @robo-pi

I don't know where Bill is headed with DB1.   I'm also not sure how much ROS will be a substitute for A.I.   I'm glued to the edge of my seat to see where Bill goes with DB1.

In Part 1 Bill writes that he is currently choosing a SBC for the intelligence layer and currently the main contender is the HiKey970.  I don't know how that compares with the Jetson Nano. I guess he will just go back and edit the text when the final decision is made.

Posted by: @casey

In Part 1 Bill writes that he is currently choosing a SBC for the intelligence layer and currently the main contender is the HiKey970

Oh no!  Not yet another SBC!  I have a limited attention span. ? 

So far I have

1. Various Arduino models
2. Two different STM32 models
3. Three different models of the Raspberry Pies (and more than one of each model)
4. And I'll soon be the "hopefully" proud owner of a Jetson Nano.

That about does it for me.   I'll be out of dough at this point.   I can't afford to keep buying every new SBC that comes out.

That's ok though, I'm not building an exact replica of DB1 anyway.   However, I just watched an epsidode of Explaining Computers where he reviews the HiKey960.   For one thing that baby costs $239.   So that's already beyond my budget.   And that's just for the SBC.  I imagine that's going to end up being a lot more by the time you start adding peripherals as well.    The Jetson Nano started out as $99, and I'm already up to $200 for a system that actually boots.    And this is already a stretch for me.   So there won't be any HiKey960 on my shopping list anytime soon.

It also doesn't appear from the review I just watched that the HiKey960 is comparable with the Jetson Nano anyway.    The Jetson Nano was actually designed with A.I. in mind and NIVIDIA is currently one of the leaders in this field.  So the Jetson Nano actually looks more attractive to me than the HiKey960.   I don't see where the higher price for the latter is justified.

Also, the HiKey960 appears to be designed to run Andriod?   I don't know anything about Andriod and that would be just one more thing I'd need to learn about that I would really prefer not to have to bother with.

So I'll just sit back and watch where all this goes.   I still haven't ordered the Jetson Nano yet, but I'm getting more confident every day that this is the correct path for my purposes.   I was a little leery about the Jetson Nano in the beginning because I read where others were having a lot of problems with it, including problems with the OS.   But based on some videos I've seen more recently including those of Paul McWhorter I'm thinking this may not be as bad as others have reported.   Especially since I won't be running ROS which is what a lot of people were having problems with.

Posted by: @spyder

Posted by: @robo-pi

The extra space needed for the Jetbot software may have a lot to do with ROS. 

Precisely

That's why my other 2 cards can be only 64g, while the Jetbot is 128. It's the ROS that's taking up most of the space

I've decided to go with buying two 64GB cards to start with.   Because I like to back systems up once I go through the trouble of getting everything all set up, and it's a lot easier to back up 64GB than 128GB.

I can always move up to 128GB later if I find that I really need it.   I'm not planning on using  ROS, at least at this point.   I'll only go to ROS at a later date if I see where it can really do things that I couldn't otherwise do.  Thus far I haven't seen any examples of that.