Building an AI Robot that can be trained! || Using an NVIDIA single board computer

[@tafsirnahian669]

NEXT EPISODE: HEART PACEMAKER 'DIY' or 'BUY'?

[@elissitdesign]

🤪

[@drobotk]

Yes, and of course, with an arduino nano.

[@joewulf7378]

sounds more like 'DIE' or 'BUY'

[@hansdampf640]

@@joewulf7378 it´s fine electronics but no nanoscience haha

[@JjMn1000]

Yes

[@NullPointer]

The thing about AI is that it's not creating an algorithm to compare the current image to the ones you fed it before, but instead creating a model that can generalize what "blocked" and "free" means, even for images or places that it has never seen before, that's the awesome thing with AI

[@greatscottlab]

Well explained :-)

[@BlackXeno]

If I expect an AI bot, it would be to recognize what blocked and free means based on certain parameters, rather than a pattern in an image associated with blocked or free (in this case the leg of the chair might mean blocked). so I assume using it in another room with different walls or chairs ecc would completely confuse the bot. did I got the concept right?

[@andrekoczka3777]

@@BlackXeno not necessarily, if you feed it with tens or even hundreds of thousands of images from houses, it would pretty much work in any room. In this case, if the other room has similar walls and general traits it might still work, just not as well as in the original room.

[@NullPointer]

​@@BlackXeno something like that, let me explain further. Old image recognition methods will have the problem that you mentioned, because they were based of examples of concepts, not concepts themselves. Imagine that you go to a new place, and there's a chair there, you have never seen that chair before, nor have you ever seen that kind of chair, but you know what a chair looks like, so you don't have to be told what it is. AI is the same, you teach the AI to know the essense of your problem, object, etc. Imagine that you want your AI to tell wheter something is a chair or not, so you feed 5 images of red chairs, and 5 images of green "not chair" objects; your data will not be varied enough to tell what a chair is, so it'll try to guess by very limited patterns, so for example, because it always sees photos of red chairs, it will assume that all chairs are red, so green chairs will not be recognized. The key is in training you AI with very varied and clear data, so it starts to recognize your object by it's core concept, and not what differentiates two similar kind of objects. In conclusion, well trained AIs will generalize, that means understanding your problem well enough that they can recognize patterns on data never seen before, just like we do. I hope I was clear enough! :)

[@BlackXeno]

Thanks to all. there's a way to ask to the network if it got the right concept? I read once of a tank recognition AI, failed, because associated the background grass to be the enemy. Which proofs the need to truly varied data. I wonder if was possible to verify this ahead, analitically...

[@syedsulaiman8380]

Never clicked faster before on a great Scott video

[@usmanmehmood55]

Same

[@devminsubasinghe6380]

Sammee

[@greatscottlab]

Awesome :-) Thanks

[@MrCytrus]

Same my dude ;D

[@tanmay______]

Maybe the next step could be adding another camera for 3D vision.. Could the Jetson board handle that?

[@greatscottlab]

Should be able to pull it off :-)

[@sarmadrafique4472]

what about a ToF(time of flight) camera

[@GCKteamKrispy]

@@greatscottlab Or lidar for 3d vision

[@mubashirsoomro6]

Can attest to this, yes it can, as long as the stream is 720p@30fps, but even then your frame rate will be close to 1 FPS or maybe a bit higher Edit: Just to add to this, the board is not powerful enough to handle ML based inference and 3d vision at the same time

[@BenjaminAdkins97]

You could look into the Intel RealSense Depth and SLAM(location) Cameras. They do some of the computation on the camera themselves so it isn't as taxing on the Jetson nano. JetsonHacks is the best channel for that. He's made lots of videos specifically about Jetson + RealSense cameras and even manages a few git repositories, really good stuff.

[@reecejones5451]

This could be a really neat unsupervised learning project!

[@sieuweelferink6852]

Actually it is supervised learning since a human manually annotates the images into different classes. In this case either obstacle or no obstacle. Unsupervised would be if both classes would be shot randomly and not ordered by a human into separate folders. And then let the AI model figure out a pattern by itself. For example a VAE would do this.

[@georgemazzeo7226]

I think he just meant it could be. That’s what I was thinking too doing it with reinforcement learning instead and let it bump into a few walls. You probably could get away without adding another sensor if you can see that your trying to move forward but aren’t but another sensor would def make it better for detecting when it hit a wall.

[@sieuweelferink6852]

@@georgemazzeo7226 could be something to try. But reinforcement learning requires a lot of trail and error. I am talking about a couple thousand tries to get something driving. Let alone advanced navigation. So you need to physically put the car back at the start a couple thousand times which is not really doable.

[@reecejones5451]

@@sieuweelferink6852 I was thinking you could add a bump sensor on the front, if the car hits something then it could automatically back up and maybe choose a random direction to turn (perhaps weighted by its confidence in the blockedness of that direction)

[@Enigma758]

If the robot had an ultrasonic sensor, it seems as though it could take its own pictures and learn on it's own without having to be explictly trained.

[@ridinggoose4169]

Humans: *AI exists* we are going to die AI: if picture equals to "blocked" turn, else drive

[@TheJay6621]

After reading your comment, GPT-3 laughed in ML

[@rpavlik1]

Cool! Note that the raspberry pi, especially pi 4 can do deep learning tasks: Adafruit has a bunch of demos showing it off. But, I think they all use tensor flow lite, not pytorch.

[@baldevmakwana5880]

Please prepare a video on ,'multi modal mobility morphobot project robot'

[@alexwolfeboy]

A cool way to continue the project, is implementing a pathfinding algorithm. Use the object avoidance to have the robot navigate to a specific location within your house.

[@MichalKottman]

"Python... the language used by Raspberry Pi"... Sorry for being pedantic, but Raspberry Pi can run any compiled or interpreted language that has ARM compiler just fine. C, C++, Java, Go, Python, Lua, Perl, Nim and many more run just fine.

[@k0r1n50]

I like the fact that you face the camera to urself when doing the intro

[@greatscottlab]

Thanks :-)

[@osa1121-]

U know!!?? U should make a car design smarter than Tesla That stops on all stop needs And so on!! I am lazy right now as it's 9:43 in the night so there are grammatical mistakes so just ignore them ;)

[@xboxgamer9216]

U sure u r not also drunk.

[@osa1121-]

@@xboxgamer9216 idk I was just lazy :) So i wrote what came to my mind

[@jakob1379]

Really interesting video, and I would for you to continue educating people with interesting idea. Can we please agree to not call it AI? It's a meaningless term that should be replaced with proper descriptions such as machine learning, computer vision etc. In the appropriate situations

[@jamesgrim8606]

From his first video to all of his videos always a joy to watch and learn something new. Keep up the awesome work you do Great Scott

[@greatscottlab]

Thanks, will do!

[@powertomato]

The creation of the model/neural network usually takes quite a bit more computing power than the network will then take once deployed I wonder if it's possible to train on a computer with a powerful GPU and then deploy the model on something rather weak like a rapsberry PI

[@favouremmanuel8876]

I think ultrasonic sensor could be used with camera sensors to make the Robot notice objects and obstacles more efficiently

[@pedrovelazquez138]

I recommend the courses from Coursera on Deep Learning. They are really helpful! At first I did not understood how it worked but, with the courses I realized the tremendous potencial that it has for many aplications. It merges nicely with mechatronics and electronics too.

[@iAmCbasBoy]

Do you recommend a specific one? I just started looking into AI CV and have some python experience

[@pedrovelazquez138]

@@iAmCbasBoy You can choose the Courses from Deep Learning with Andrew Ng. This is with Python. But, if you want to know the details about machine learning and neural networks, you can choose the course Machine Learning with Andrew Ng from Stanford... the last one is free.

[@69iqtutorial]

Uh i dont know why im so late, i got the notification 10 min later, any solution? Pls tell if you know...

[@greatscottlab]

Turn on all notifiactions :-)

[@69iqtutorial]

@@greatscottlab actually all are turned on

[@BoingoInstaller]

Try to unsubscribe, wait a minute, subscribe again and turn on notification KZbin is truly piece of "you know", we should move to another platforms, LBRY for example

[@ebrocoliphoto]

Jeremy : put nvidia on title PC guy : hello

[@Produkt_R]

Let's see how it reacts to a mirror

[@Ltsoftware3139]

Cool! It's amazing that people from all domains get interested in Machine Learning. Every domain of activity can benefit greatly from using these algorithms, but sometimes we underestimate the domain-specific knowledge needed to understand and solve the problem.

[@greatscottlab]

Absolutely!

[@ollimacp]

Deep Learning is totally possible on raspi 3 or 4 (even convolutional neural nets), when training the code on a pc -> just search for Donkey Car and you will find everything you'll need to know. Expensive Training needs a GPU just running the network doesn't need a gpu.

[@MCsCreations]

Really interesting indeed! 😃 Too bad it can't learn for itself... I mean, trying, colliding and taking pictures of where not to go... You know? But other than that, it's a pretty impressive little kit! 😃 Stay safe and creative there! 🖖😊

[@electronic7979]

Excellent

[@fotmheki]

You can actually use Raspberry PI with TensorFlow Lite

[@greatscottlab]

Good to know ;-)

[@Xumatro]

1:09 "automonous", 1:47 "automonously", lmao

[@dhrubojyotighosh9592]

Im your very first viewer of this video, btw you are really a great scott

[@greatscottlab]

Hey, thanks!

[@dhrubojyotighosh9592]

@@greatscottlab Sir you are my inspiration to learn about electronics (me = 13 y/o) and you are welcome

[@knightdeluxegaming]

I really want to learned those application of machine vision but im *broke* sad noises.

[@greatscottlab]

Sorry about that

[@mohamedabdelhamed191]

You can learn them on youtube btw and implement a single digit classifier easily on your laptop, you can search for MNIST dataset classifications. Good Luck

[@sumeshsarkar9863]

For just making obstacle detecting bot, why can't we use ultrasonic or ir? It would also do the same but in easier way I guess......

[@SaminaZafar270]

Make rubik's cube solving robot which takes pictures of each side of cube and give an algorithm to solve it

[@ANTOSGARAGE]

Great video sir but I have a doubt....The same can be made with Hc-sr04 Ultrasonics sensor with Arduino which detect the obstacles right....what's the difference .... sorry I'm not teasing you but I just want to know....?

[@mr.coolio4321]

I love how Jeremy's German accent causes him to say *vikipedia*

[@revealingfacts4all]

Never knew his name was Jeremy. Assumed it was Scott all these years...

[@gregclare]

@@revealingfacts4all LOL. same. Who’s Scott then? Scott must be Jeremy’s mentor, since apparently Scott is great! :-)

[@superman111123321]

@@gregclare kzbin.info/www/bejne/r4mXi5qrhL-dntU&ab_channel=ManticoreEscapee

[@Adi-lv9bo]

This is getting absurd, every one on youtube who make an AI thing, have same stupid idea on use AI and du what can be done with logic circuits, is not even necessary for a microcontroller, just a few logic IC and some optic sensors.

[@greatscottlab]

Not quite. AI goes way deeper than logic circuits and optic sensors.

[@Adi-lv9bo]

@@greatscottlab I know what are you saying, but end result is a robot that moves randomly and avoid collision

[@Adi-lv9bo]

@@greatscottlab what about a robot that can find something using hints, for exemple recognize a pat which should follow, find piece of bread following crumbs, etc.

[@moonmatthew]

WoW! Such cool video! Thanks for next awesome video!

[@greatscottlab]

Glad you liked it!

[@lonewretch]

Holy buckets, face reveal. You are not the person I imagined all this time!! lol :) (alternative title, I've never seen a video with your face)

[@atharvap8]

Great video as name!! Surely a informing and interesting project!Looking ahead for a new video. And thanks for such a great knowledge and content. MUCH LOVE FROM INDIA😊😊

[@1NIGHTMAREGAMER]

the real reason robots cant have real emotions is u dont imedaitly have them u grow them in a 20 year span u are raised by ur parents try raising a ai for year that has no knoledge

[@wolfgangpreier9160]

I see you are very experienced in the delivery room for your children. How often have you made this experience and raised a child to know so much about them?

[@69iqtutorial]

@@wolfgangpreier9160 lol

[@666aron]

Amazing project. Maybe a balancing bot for an upgrade (wink wink)? It is fascinating how affordable these sbcs become. One thing that always bothers me, when I see people designing mobile robot project, or rather not bother, but simply curious: why do people use dc motors without quadrature encoders?

[@michalkana9764]

Almost unbelievable how today's electrical hobbyists have many ways to create fun project. Btw where you get jumper wires for your prefboard projects?

[@electronicguy4550]

Its called silverd copper wire

[@michalkana9764]

@@electronicguy4550 thank you for answer, but i also mean wire that he shown in his first essential tools video, he shown pliers that can bend this wire acros two holes in prefboard. Thick, strait wire.(i dont know how to describe it beter, im from Czechia.)

[@farahhappy4970]

keep your nvidia board in a safe place arduino boards start planning for something evil

[@vishwasshettyv]

The writing hand has a face also 😀👍

[@anddop]

If I added 2 wheels on the back, that aren't controlled by any components, would that work or would it mess up the original build.

[@peachpotatochips473]

This Channel is like my second school because I learned a lot knowledge about electronics and stuff from here. You're awesome!

[@hr.differentmind2048]

love you greatscott. . and also your voice

[@GregKeane]

does anyone have any suggestions? I have a Chinese motor driver (BLD300B) I would like to connect it to either a JETSON NANO or a laptop computer, how would I go about do this?

[@FRESNEL_COOKING_SOLAR_OVENS24]

Tesla is not a self drive. Can drive only on highways where is not coliding with other people in the junctions. Elon Musk won't manage Tesla for a very long time. As other technologies will come up Self driving vehicles can be achived as long as you have digital signals on the streets that can comunicate with all the vehicles....only when the world traffic its 100% electric. So no other vehicles can be involved or intersected from outside the wifi communication at red and green signals. There would not be human involved to control the vehicle start and stop in junctions. Because every traffic junction would be autonomously comunicating with every single lane of traffic to let cars drive and stop autonomously till reaching the presetted destination.

[@cursorop1716]

Can we do it without using raspberry Pi ... Just sending video footage to computer wirelessly and then giving it commands from computer to avoid or not.. I'm want to make project using this concept where I can just give it destination and it will automatically travel towards it avoiding obstacles

[@sarmadrafique4472]

How about using a ToF(Time of Flight) Camera?

[@kardeef33317]

A Tesla a few days ago with 1 man in the backseat and one in the passenger drove off the road into a tree and bust into flames killing both people. P.S. it's against the law to operate a motor vehicle without someone in the driver's seat. P.P.S. Here in Houston it happens alot with regular cars cause are roads are so messed up and a lot of times have conflicting markings because they have been changed and/or repaired 100's of times. If the roads here can confuse people who have driven them decades can expect a A.I. to be able to decipher them....

[@salvatorecristiano2030]

Why can't i like this video more than once?

[@tech-a-tronics650]

Would love to see a lidar connected to this bot and hopefully running ROS, we will be able to map the surroundings.

[@ahmedhesham6944]

can you make a series about jetson nano?

[@alperenkaya2798]

Firstly thanks much scott for the video. But I need to say that Ur accent is really annoying😅😳. I couldn't focus on the things except Ur voice. I'm sorry to say that but it is fact.

[@Spiritofaconure]

I would like to build an ai pet , with flying capabilities, not like a bird more a drone, I have pets that would enjoy plying with something that can learn, they love music and they are interested in everything

[@tejasjoshi781]

SIR SUGGEST ME A SENSORS WHICH CAN DETECT PLASTIC .?? Whose libraries are also available online for proteus..

[@Schjoenz]

You said "autoMOnous" instead of "autoNOmous".. it's just a small detail but I think you might need to know and will be helpful for your next video.. I love all of your videos since the first time I sub..

[@ssaniljainn]

can we implement the same Ai training model for collision avoidance in drones, too?

[@JR-je7ce]

Don't know if it was just a uk thing but does anyone remember the real robots magazines that were out in around 2001 that came with parts each week to make this little bug thing with proximity sensors and stuff

[@gaberbrkat6622]

To be completely fair you should let it run in another place other than corridor cause if you lets it run just in the corridor you wont take advantage of any AI software cause it would be just remebering photos that are already taken not comparing with new photos

[@ai_04_yadnesh_bhanu]

brother the one you made in this video is the one with Jetson Nano or without jetson nano ? Im considering this topics as my uni project so please help

[@akshitgaur8581]

When you realise a single board computer is more powerful than your pc.

[@treemanzoneskullyajan711]

Eye ball Paul could keep a pace maker in time according to Kevin at Perry's funeral.. heart failure is a bitch

[@judo-rob5197]

I read somewhere that the Raspberry Pi will in the future have an AI chip on it. That would be cool.

[@rahalrodrigo5430]

Yeah that would unlock be amazing.

[@questionmania2191]

From the thumbnail i thought it was also a self balancing robot😂

[@greatscottlab]

Not quite ;-)

[@fatonisodiq9341]

@@greatscottlab but with AI, what not to do with self balancing robot?, that would be a good idea

[@OZtwo]

Great video. I have been looking at getting a Vector Robot to play with the AI but finding it may not be up to speed. I'm simply looking for something like a Vector robot that can explore the area, charge itself, and more where the AI comes in, be able to collect small objects and take it home. For this why I picked the vector since the base design is perfect yet with too many limitations. A perfect robotic mouse which I am so shocked I can't find anything close to it.

[@mr-momomourad6475]

we need sensitive metal detector for this AI Robot ......can you make metal detector tutorial using microcontroller like arduino

[@kayakMike1000]

I am totally going to replicate this project and share my results!

[@Nway257]

I'm 16 year old and I have a wish for learn anything that has to do with artificial intelligence and hardware electric engine ect .....

[@gameandelectronic8369]

Hey, can you make a video, about making a simple code for tft display, so it can play a video and open some image. I already try it l, and i can't make it work

[@ThorstenDeuter]

You actually made a product video but I must admit a very good one :D

[@igen2019]

Good... Good.. But this kit is very expensive. (250 euros)

[@upekshawetthasinghe906]

Yo you even don't know about the graphic cards Nvidia make , you have no idea about computers

[@ajaymote4950]

you never disappoint me, your the best😁😁😁

[@sandeepnaik6818]

*Give it shape of R2D2*

[@emmoemminghaus6455]

omg... there is no artificial intelligence... even when a blone trys to dye hairs black... there are only trys nothing else

[@DanielH]

Now teach it to take over the world

[@kiniax2753]

i seen the prices and decided i cant afford to even open youtube videos abt robotics let alone actually start the hobby ouch

[@JCBOOMog]

Just got home from a walk

[@greatscottlab]

Nice :-)

[@blackerhawk1508]

Thas an awesome pice of hardware, but, I have a question: can it run crysis?

[@ishigamiyu1991]

Automonous vehicle

[@PhG1961]

Finally... what I've been waiting for !

[@tmastersat]

AI my butt. Why is everyone claiming ai now this 8s just programming not ai

[@ICE__T]

I want to create an ai robot for the plumbing industry, where do I start?

[@samda7109]

Schönes kleines Spielzeug. Mit IR-, US- oder Radar-Sensoren hätte man sich allerdings das teure Board sparen können uns selbe Ergebnisse bekommen. Für AI mit Kamera ist das aber echt Nice geworden. Erinnert entfernt an die neuen Drohnen die Objekte und Personen verfolgen können ohne gegen Hindernisse zu fliegen. www.drohnenhandbuch.de/follow-me-modus

[@anoniem012]

Why did you show yourself, you wad always anonimous now you have broken it

[@lambsauce5445]

I just got into PyTorch and saw this. Perfect timing. Thanks.

[@michaelmullen7237]

It been a long time

[@delvinthomas]

Super bro 🙏

[@SEEMAKUMARI-qf3qk]

Simply use Arduino and a ultrasonic sensor. 😂😂

[@emorobotturkiye3639]

Can we add a speaker and a microphone and add an artificial intelligence that can talk?

[@ArabicSalamander]

Change the title to "How to assemble Pre Build AI robot set"

[@wafiullah-shafia]

So much great idea Good explanation Thank you

[@greatscottlab]

Glad you liked it!

[@userqwertyuiop9652]

You never heart your subscriber's comments

[@Roe124]

I am new to programming. Is this considered hard to do?

[@gillianschalenbourg5867]

You Should re do this project as Ai has come a long way…

[@sanfinity_]

Last week i though of suggesting you with an ai based robot sir but to my suprise it actually comes true😂😂 keep making this kind of videos 👍🏻🔥

[@sanfinity_]

Also try using ROS(robot operating system) sir.

[@greatscottlab]

Thank you so much 😀

[@Un_Pour_Tous]

It's not a Unitree however a pistol straped to it will give it a chance.

[@pitaco-0815]

Danke. Elektor Heft ist bestellt :)

[@mj2068]

constructive dislike for not using desktop computers to train.