NVIDIA JetBot: Jetson Nano Vision-Controlled AI Robot

[@digitalghosts4599]

Jetson nano is a ridiculously powerful platform for the price. I'm using it for high speed machine vision to monitor manufacturing process in real time and this beauty can record 200fps in 720p and it can easily process 100 frames per second with a simple detection algorithm. We are living in the future. 10 years ago this would be a huge challenge even for a desktop PC not to mention that there was nothing in place to capture such high frame rates in real time and process them simultaneously.

[@ExplainingComputers]

Great feedback, thanks for sharing. I share your appreciation of the Jetson Nano -- it is a really great board.

[@travelstories7530]

how do u make that simple detection algorithm is is based on those green square u defined in the algorithm?

[@hussainbharmal5998]

How is it going on, after 3 years? Have you been using it to date in other applicationa too?

[@JohnK68]

Actually it's not powerful at all. A nvidia tegra T210 with a GM20B GPU it's a toy. Also the OS on the board is way to heavy for the system.

[@chriholt]

All I can say is "Wow!" Very impressive hardware and software package!

[@gpalmerify]

This video helped me appreciate my Subaru's "Eyesight" system even more. Thank you Chris.

[@MegaZiggo]

This is quite interesting. I worked as a Field Engineer for a company call Intelligent Reasoning System Inc. back in 2001/2002. They made capital equipment for installation on an electronics manufacturing line for use in Automated Optical Inspection of PCBs. We used an early form of image training (the founders of the company also wrote the software for the terrain tracking algorithms for the Tomahawk cruise missile and adapted those to this application) with statistical analysis within a set standard deviation. The system worked well and it is very similar to what you demonstrated here with the Jetbot. We would train examples of good images and bad images of electronic components on a PCB and for each false call, we would train that as a bad actor. Over the course of the manufacturing run, the system would become more and more accurate. Good stuff indeed. Keep the good material coming! I am considering getting one of these to teach my daughters not only Python, but machine learning in general...

[@johncnorris]

nVidia: "It just works!" Explaining Computers: "After extensive assembly and configuration it does."

[@ExplainingComputers]

So true. NVIDIA were very helpful during production of this video, but production lasted longer than on many other videos. :)

[@jasontiscione1741]

​@@ExplainingComputers Probably starting when you first plug it in and Ubuntu thinks you have a black and white TV

[@lollerich]

Very, very interesting. I finally got at least a general idea of how neural networks are trained. Thank you!

[@ExplainingComputers]

Great to hear this -- that is what I was trying to convey. There is a fair bit of process between gathering the sample data and having a final trained model, but the essence is what you see here.

[@lollerich]

@@ExplainingComputers Thank you. While I did in general understand the purpose of neural networks I was always very hazy on how you would actually train one.

[@noealves8400]

I have a nano sitting on my desk for a month or so now... finally I found something cool to build with it! thanks for the video!

[@twmbarlwmstar]

I have about 16 SBC sitting under my desk, so I laid off the Jetson until I'm mature enough to go on eBay unsupervised.

[@62shalaka]

Sundays are exciting; I never know what Chris will present to us next. Great as usual!

[@phildodd9942]

A useful presentation you have given us, pointing out that AI can be other things apart from analysing numbers or text ! In fact you've made us sit up and take notice ! So THAT'S how a self-driving car parks itself ! Thank you for this enlightening demo !

[@lowiehojas5525]

I was just leisurely watching, but instead I learnt something that I need for my school assignment :D I just learned from your video how to train a data set for visual detection! Thank you so much!!

[@rodlupanow69]

Amazing !! I bought the Nano today and I am very excited to start experimenting with Deep Learning on robots. This is a very motivating video. Thank you.

[@ExplainingComputers]

Good luck with your Nano. It is an amazing board. AI is tough to get into on any platform, but Jetson has better support and resources than anything else I've come across.

[@foobar1231]

Congratulations on your first trained neural network!

[@PeteVanDemark]

Fascinating! Nice little robot. Never would have guessed it uses an image library to navigate. Like the table too!

[@ExplainingComputers]

Every ExplainingComputers video where I look at any hardware is shot on that table. :)

[@arthurmoore1379]

@@ExplainingComputers My good man this is way to complicated for mm. Arthur here:

[@Osmanity]

Totally mindblowing video! I just wish that Nividia sold pre-assembled JN-AI Robots versions so that we can just focus on building the software and learning more about training a neural network which is for me the main goal. Thanks as always for an interesting video!:0

[@ExplainingComputers]

Since I made this video there are some suppliers selling kits, which makes things somewhat easier. I've added the link to the video description.

[@Osmanity]

@@ExplainingComputers thanks

[@iluvrgb]

Back again with another interesting video. The Nvidia Jetson Nano is interesting

[@AnthonyCook78]

What a waste of time!

[@chadwick2629]

@@AnthonyCook78 How?

[@Bippy55]

This is one of your BEST videos. I'd favor the company creating a kit of parts. Like Heathkit used to. Then you'd know the parts should work after assembly. I'd also like a BEGINNER's PRIMER on programming the neural net robot. But overall, "Bravo!"

[@ExplainingComputers]

There are now kits -- see links in video description! :) Glad you like this video. It is not one of my most popular. :(

[@ericartman0]

Since watching your wonderful channel my "want to do " list never ends. Thank You!

[@ExplainingComputers]

Thanks Ted. For anybody seeking a complex maker project with a lot of possibilities, the JetBot is really interesting. And I'm sure we are going to see more and more SBC-related machine learning and robotics stuff.

[@bosstroll9019]

If Mr. Scissors found a way to mate with it and breed, they would rule Earth by Thursday

[@VeryUsMumblings]

1st choice: Ruke the world. 2nd choice: make lots of unboxing videos!

[@brianm6337]

Meh- they can have the earth. Kids will be a bunch of little cut-ups, though.

[@brianm6337]

@b gg Explaining Computers does. ;D

[@cinnabarsonar2072]

Gives a whole new meaning to the term "hardware porn" I'll grab my coat.

[@bijanshadnia3620]

I thoroughly enjoy your videos. This one in particular is one of the coolest I've seen on your channel. Thank you for the hard work you do.

[@MichelMorinMontreal]

"Bring me to your leader...!" All right, all right.... We're not there yet with the NVIDIA JetBot! Quite interesting prospects, though. Thank you for the educational presentation!

[@NewAgeDIY]

A great class room project platform. Computer science lab now has a low cost system to work with. Thanks for sharing this wonderful SBC computer Chris.

[@ExplainingComputers]

I totally agree Dale. The JetBot is a fantastic learning tool -- for all ages!

[@arashi926]

Alright, this is a bit more challenging. Definitely have to try this one. More of this please! Keep those great videos coming. Really enjoy them.

[@ExplainingComputers]

Thanks for this! Sadly this video has not been that popular . . . :(

[@arashi926]

@@ExplainingComputers Sorry to hear that. But it doesn't surprise me. This looks like a quite advanced project. Not for the a broad audience like an Arduino (or at least, that is what I think is the reason). Anyway, happy you took to time to share this.

[@southsidedojo2]

Chis i got your coffee mug and added it to my collection! very nice and proud to display it!

[@ExplainingComputers]

Great to hear. :) Thanks for being part of the EC community.

[@CodyBanks10]

"Something like that" is a buzz phrase I will be using from now on.

[@j2simpso]

Thanks for sharing the project Chris. I sure hope Nvidia makes these kits available for educators as this really is the Lego Mindstorms of the 21st century!

[@gplayer01]

Excellent review & demo Chris of the Jetson. Interesting to know the amount of pictures you took to train the Jetson. Well done

[@rogerkoh1979]

Well technology has come a long way. I remember a robot doing obstacles course about 20 years ago. It took a long time just to move. The ultrasound sensor is a lot faster back than. Now it has catch up. Great video as always.

[@World_Theory]

A while back, I got an Idea… This is how it goes… You have an Artificial Neural Network (ANN), that is having trouble with a task that requires a perception of Time. What is a simple way to give it a perception of time? Answer: Give it memory. What is a simple way to give it memory? Answer: Use data from its sensors, taken at consistent intervals, and give them a “memory slot” in a cache. Use a small number of memory slots for progressively older sensory data samples. Whenever a new sample is added, delete the oldest sample, and shift the contents of all the other slots to the slots representing the sample that is one interval older than them, until the slot representing the newest data sample is empty, so that the most recent sample can occupy that empty slot. Feed the ANN from whatever is in the memory slots, as a separate sensory input. It will then be able to compare old vs new, and perceive changes from as far back as the memory slots go. This will need careful optimization, to avoid overwhelming the memory capacity of the hardware that it's all being run on, and to not require a ridiculous number of “neurons” in the ANN to be dedicated to input. One method of optimization, may be to also perform a resampling operation on saved sensory data, to step it down in size, every time the data is shifted to a new slot, whether immediately after it first enters memory, or if it meets the condition of being past a certain age, so that it loses fidelity, but still exists for reference for much longer within a given memory size constraint. Some care should be given to choosing a good resampling filter. It should probably be a balance between accurate and lightweight for processing. And accuracy brings up another subject: Resampling images while stored in the sRGB color space, is mathematically proven to distorting the overall energy of an image. All camera input should be converted to linearRGB colorspace, if it's going to be subjected to resampling. sRGB is meant to make efficient use of digital memory, for cases where an image is meant to be shown to humans, by matching assigning more importance to ranges of the color spectrum that the human eye is most sensitive to perceives differences in. So sRGB is completely unnecessary for computer vision. Another optimization method is to simply start with a smaller sized resolution for sensory input, or to use a color model other than RGB, such as Lab, where there is one color channel for Lightness, and two color channels for chromatic data. Then, you can aggressively down-sample the chroma channels by a positive whole number. I've mentioned resampling a lot, but I think I should mention that I'm not 100% sure how useful it would be to save processing power overall. I'm quite certain that it will save memory, though.

[@bobpaugh6725]

Mr.Barnatt another great video. I learn so much watching your videos. Thank you.

[@ronjenkins4257]

I can imagine networked robots of this sort, each perhaps in a different school, racing around a common virtual racetrack or in battling in a common virtual battlefield. In actuality each school's robot would be operating on an agreed upon real physical space (a school gymnasium), but playing on a common virtual space, populated by virtual robots being generated by the physical robots in each location. The winning robots would be the ones with the best-programmed algorithms supplied by the students.

[@smartassist9700]

Thank you yet again! THAT IS PART OF MY "KING PROJECT". i want to automate lawn tractor mower to mow lawn using visual and distance sensor for obstacles but program to mow in a pattern. Learning properly to go around objects on eventually both sides to keep mow pattern. One thing I have not figured is to read where grass has been cut verses the next cutting row of taller grass. Unless it uses "mapping" and location to determine the next row. That would make sense. The servo motors for steering and such I have to figure out with experts in that area. There is one company that has pattern mowing already. I am discussing selling "kits" to install on older mowers already purchased. They seem interested. but that will have to wait as I have to many automation things to complete first. Also paint a workshop barn/office. Insulate, run wiring, and interior walls and set up work table, hand tool storage organized on wall. (I am building skeletal arm that requires servo motors camera and nvidia to run it. Helper Arm to pull tools and put them up when finished. I have 4 months to also setup garden, water retention, sensors pumps water lines. This year with no garden I grew the most delicious Pineapples, Lemon grass, tomatoes, and seasoning herbs...playing around. I need some automation interested people to assist with this project. amazing the blue print for automation has grown substantially. I guess it will take 2-3yrs to implement fully. I would prefer that 24hrs a day over any other option to spend time on.

[@Peter_Enis]

Use bluetooth adapters in the corners of the garden as "gps-beacons". Now you can use the mower to print text or pictures in your lawn like a big plotter.....?

[@NataLia-yb6vm]

Now I no longer need a Tesla for self driving car. I can make my own haha

[@Colin_Ames]

Thanks for another interesting video. I always learn something new after watching an episode of Explaining Computers.

[@twmbarlwmstar]

Impressive, I managed to write a simple IA script this week, as in copy it and get it to run, and that was a real first. It didn’t really do anything as such, but I could see the code working and that helped me get my head around things. Really, I struggle with all this and always will but just making some sense of it helped. My 10-year-old son will hopefully do better and having something like a robot will be more meaningful than my naming game. A £100 price point seems about right for this, especially as it should be extensible and it is the code that is what it’s all about, and you need affordable so there's access. Hopefully you can have say a team of 4 kids with one Jetson to build a project. £25 each isn't a complete killer over a year, say £1 a week- even those on very low incomes should be able to support that. To be honest though, I’ll always be a hardware man at heart so the Jetson is a bit lost on me, I’ll only turn it into a Kodi box or games emulator. EDIT: I do wonder what Raspberry will come up with, clearly it won't be Jetson standard as Nvidia are miles ahead on this stuff and have massive budgets. The same for AMD, I just wonder there, they have supposedly invested some money in IoT/Embedded but I think they will be in a very different direction- could be wrong but more 86 64 and industry utility stuff- certainly not self driving cars.

[@BlenderRookie]

Wow, I wanna play with that. Plus the commands seem rather intuitive. The commands kinda remind me of that old graphics program called Logo Writer.

[@elviraeloramilosic9813]

Hello Chris👋🏻 Great job you did here. Amazing little robot! I recently included ML and NN in my learning schedule, so this is most beautiful example I've seen and a excellent ground for tinkering/experimenting/learning! Fascinating!

[@ExplainingComputers]

Hi Elvira, it is always good to hear from your here. :) I too have it on my list to learn more and more about ML and NNs. It is so different from traditional computing! Sadly, on the channel, it is not that popular a topic to cover. But I will persist on occasions like this one.

[@elviraeloramilosic9813]

ExplainingComputers Yes! Please do cover this topic. 🤩 At least occasionally. 😁

[@swapnilwabale2227]

Hi what job u r doing

[@swapnilwabale2227]

Elvira what job u r doing

[@SniperUSMC]

This would be a good project to use to train an autonomous lawn mower. The direction changes need to be in a bit smaller increments so it doesn't just turn too much at once but just a little and retest for obstacle. If all the information can be stored on SanDisk card and saved then the trained brains/program can be moved to a bigger robot and have the "Blades of Death" (lawn mower blades) installed and turned loose safely to mow the lawn.

[@suvetar]

That's awesome! Thanks for making my Sunday again!

[@stanislavkotzev4157]

7 years later he hasn't changed at all :D just watched the big data video

[@superdupergrover9857]

I was hoping for a jet-powered robot, but I suppose making machine vision robots easier to use is pretty cool... (btw, using a powerbank as the battery is a really smart idea, makes battery management easier for newbs)

[@KangJangkrik]

This old man never get old

[@spidermcgavenport8767]

Thank you, Mr. Barnatt, Explaining Computers, My jet bot would be using vintage fisher price contrux.

[@spidermcgavenport8767]

Short Circuit S.a.i.n.t robot Johnny 5!

[@MacPhantom]

Great to see that NVIDIA uses PyTorch!

[@Techn0man1ac]

It's alive, it's aliiiive

[@michelfilion5482]

Amazing...If anything, AI shows us how we take for granted our own complex cognitive abilities.

[@maxvaistuk5493]

www.3dmaxprinter.com/shop/do-it-youself/rc-robot-car-with-ultrasonic-sensor-for-education-robotics-robotic-car/

[@GregSilverado]

years from now our robot overlords are going to remember your kindness to one of their little ones.... they will be merciful to you

[@ExplainingComputers]

:)

[@NomadicSage]

This was a great video, please do more videos on AI and programming

[@ExplainingComputers]

Thanks. I hope to return to AI, but so far -- and after many attempts -- it has never really taken off on this channel.

[@NomadicSage]

@@ExplainingComputers it's quite different and interesting to see it in action, the experiments that you do in your style of breaking them down into easy to understand components is what always draws me to watch your videos. As well as replicate them with my own configuration.

[@twmbarlwmstar]

@@ExplainingComputers I want EDGE! No seriously, after you teased EDGE I've been reading up on it and looking forward to your planned video, I an see a direct use for EDGE computing in my life more so than IA, despite things like Amazon Dot or what have you. I think IA is a hard sell full stop, because of people like me, I get the concept but I don't get the access- as in I'm a bit thick and just don't get it and don't have the coding skills to get it. It won't go down well but I'm finding some of the MS stuff a bit easier than the Linux stuff- with the proviso I am on about stuff targeted at complete novices. And I'm not making any claims that MS W10 embedded is in anyway better, I just found some of it more accessible to me- probably because you get W10 as a base to build on. And all it is a doorbell that can tell you if it is a bear at the door or it isn't a bear at the door and that's it. There aren't many bears at my door (a few wolves maybe).

[@NicoDsSBCs]

I hadn't seen this video before. I must have been on a trip then. Great video as always. Greetings.

[@ExplainingComputers]

Hi Nico! :)

[@MrVein5.0]

wait a second, i got a 1hour ad before this video? wow. interesting stuff but wow.

[@smartassist9700]

I am finding sponsors! The used mower I purchased. The brand Company has not produced in 7+ years. However, without asking they shipped me many new parts they had stored for that model mower. I will purchase the remaining 3-4 small items from them. Mower will be better than NEW.

[@willyarma_uk]

This is very cool! Now can you program it to search for a rabbit?

[@stevehallam0850]

Fascinating and well presented. Thanks Chris.

[@TheTwick]

Thank you for this video. The definitive test would be to train it to follow the voice command : “Robot, get me a beer!”

[@Avionics1958]

Truly fascinating to watch! Thank you.

[@perrymcclusky4695]

Very Cool! Looking forward to your next video.

[@ExplainingComputers]

Thanks Perry.

[@williama29]

I like robots and AI I wouldn't mind having Mr. Scissors as a AI robot

[@BharatMohanty]

I like this video sir, very informative..I made a terminal controlled rover but I without microcontroller or SBC ..

[@areceemaz]

I showed my wife your "Ryzen Budget PC Build #6"... Dinged my wallet, but she's thrilled with her new system :) My wife and Amazon thank you, LOL

[@ExplainingComputers]

Great to hear.

[@thomasottvideos]

Tesla Model 3? We don't need no stinkin' Tesla Model 3. We've got JetBot!!

[@cocosloan3748]

In a 20 years one neural network talks to another.. -Today i trained my first human :)

[@ExplainingComputers]

Such a powerful idea and post. Really like this.

[@Flix-f6q]

Don't wait 20 years for that. Considering that I use KZbin videos to learn about Arduino and Esp32 microcontrollers and that it occurs frequently that suggestions pop up in my recommendations that are really interesting for me because they cover the same topic, I argue that we are already somewhat there. So far, it brought me from basic handling of GPIO to Machine learning and AI.

[@adaemus333]

Good robot. Robot move only straight and turn left. For this hardware its not poor. Again nice video.

[@ExplainingComputers]

The robot can move/turn as required, the code here just says "straight forward or turn left if the neural net reports blocked".

[@jyvben1520]

the birth of the jetbot whisperer ;-)

[@0dyss3us51]

What I would suggest is to put a sensor under it where it can detect edges, I have a robot made years ago that goes to the edges then stops because of the edge sensor it works like a charm!

[@ExplainingComputers]

But the point here is a demonstration of a vision recognition neural net that can be taught. :)

[@0dyss3us51]

@@ExplainingComputers right :) I guess both does the trick!

[@MicrobyteAlan]

Fascinating. Absolutely Fabulous. Thanks

[@arthurdent8091]

Hi Chris. Nice video. I caught the pun EYE instead of I, Robot, nice one. Robot vision will interest me more when I can be inside a self-driven car. Cheers.

[@CRTECHPE1]

nice one. after 5 year, those things can be much easier

[@0dyss3us51]

Would love to see you expand on this robots capabilities! :D how about a hand that can grab drinks from the fridge? Okay a little ambitious, buuut possible I guess :D

[@abfig78]

Love it!! Thank you for your time!!

[@srtcsb]

Very cool Chris. Of course it's 'simple', stuff like this starts out simple (you Arduino & Raspberry Pi blinky lights folks will know what I'm talking about... You can learn a lot from blinking those lights! ). Does Nvidia show you (or give access to) the code for this? Like another commenter said, possibly you could modify or extend the code for more functionality. I'd love to build one of these and examine the possibilities. Thanks for another great video Chris.

[@ExplainingComputers]

Hi Steve. And yes, all the code is available to tinker with. And there are various other examples available, like self-driving down a road based on its markings, and following a person/object.

[@saturno_tv]

Greetings Chris!. Expecting some review about the up-board.

[@ExplainingComputers]

If they send me one . . . :) Their new stuff does look very good: up-board.org/

[@dancingCamels]

Would be interesting to see the least number of training data images it takes to work effectively.

[@qzorn4440]

wow very nice project. 🥳 I sure miss the old Heath-Kit days. When it was a turn-key project with everything in a box and a great manual. Need a Teach-Kit company to replace Heath-Kit?

[@Sam_Saraguy]

Move aside laser pointer, the cat has a new toy.

[@andraugust4084]

Hi Christopher, Thanks for the great video. Did you find the bot to be energy efficient? Or did running the NNs drain the battery quickly?

[@ExplainingComputers]

The battery lasted really well -- JetBot can run for periods of many hours on one charge.

[@didiyontingwi]

Interesting.. I like this product.. Thanks.. Greetings from indonesia..

[@ExplainingComputers]

Greetings back from the UK. :)

[@statorworksrobotics9838]

Another fantastic video but perhaps it helps to highlight the problems of current AI approaches. Bandwidth and data hungry, non transferable knowledge. Whereas a simple symbol based heuristic solution would probably suffice.

[@waynerobarge8543]

Thank you for doing this video. I have played with the nano but have not looked at the robot feature. Question: the robot seemed to favor turning in one direction for avoidance most of the time. Is that a random variable - like flipping a coin - or is the robot taking additional information before making the decision which way to turn?

[@ExplainingComputers]

The robot is simply following the rule "if neural net reports blocked, turn left".

[@roosboom]

Great video. Thanks for sharing!

[@CTCTraining1]

I fell happier to know my Tesla is unlikely to fall off the table.

[@neelstheron3252]

This can keep you busy for hours! This is a nice toy (with an excuse) in the name of science!

[@cvicracer]

great channel, i love all your work, thank you!!

[@ExplainingComputers]

Thanks for watching!

[@reggiep75]

I would train this to hunt my daughter when after she's destroyed some of my buildings on Minecraft. But on a more serious note, this was the kind of stuff I wished to see as far back as the 80's having witnessed lots of projects on technology TV programmes and wondered when they'd arrive and under what level of control they could offer. I could imagine the training process being a laborious task unless there were numerous libraries/databases you could call on and modify which I suspect their will be over time.

[@ExplainingComputers]

You are right -- there are already vast pre-trained neural nets being built up -- like the one I showed in my last Jetson Nano vision recognition video, where it was identifying objects.

[@Azureskies01]

another amazing video!

[@johnsweda2999]

Instead of having a camera to measure edges wouldn't it be better to have an optic measurement, you could use form a mouse, so it can count how many steps / rotations it's doing and work out the area surface and produce a map. This could be used in conjunction with the camera, I mean a camera doesn't know the area it's patrolling with a counting system it is easier for it to work it out I would have thought

[@ExplainingComputers]

I am sure that you are right that using/adding sensors other than vision would improve matters. :)

[@Solidahmed]

Great video as usual. Thank you for sharing and explaining the whole operation of this demonstration. I noticed that the machine turn in one direction only. Is this usually the case? Or can it choose to go the other direction if needed?

[@ExplainingComputers]

It turns one way just because the sample code here just says "straight forward, or turn left if the neural net reports blocked".

[@ryancoke777]

That roller ball on the bottom reminds me of an old mouse ball. That leads me to a great idea: a Jetson nano controlled mouse using a Bluetooth gamepad. Useless but why not?

[@ExplainingComputers]

I'm sure your idea could be done. :)

[@bradscott3165]

Jetbot should be named Dale because he only turns left.

[@ma-burke]

(Blib #6336 in the EC digiverse.) "...to see what it can do. So let's go and take..." _time passes_ _much time passes_ _years in fact_ "... a closer look."

[@oundhakar]

Absolutely fascinating. How can you prevent it from getting into a loop? Maybe switch between turning left and right alternately?

[@ExplainingComputers]

Some randomness in its turn direction would probably be the answer as you suggest. I find it fascinating how an "optimal" pattern soon emerges.

[@1974UTuber]

Hi Chris. Another great and interesting video. Any idea why the robot only turned Left in collision avoid mode? Seems like it was only made to avoid by going Left

[@ExplainingComputers]

Yes, the code is set to move forward if the neural net reports free, turn left if not.

[@angelleal3005]

All those packages & libraries on the jupyter notebook have to be previously installed on the local machine( desktop computer ) ? I'm quite a newbie on this topic.

[@mtbevins]

Great video. I wonder if you could fit a PIXY camera on the Devastator chassis and so something similar. The PIXY camera can also be trained to recognize specific items and colors. I have thought to build a robot based on one you have built in the past that drives around and seeks out scorpions around my house. Sound like science fiction but with the available single board computers and hardware I think it is very doable.

[@ExplainingComputers]

I love the idea of a scorpion-seeking SBC robot! :)

[@TheDavidPoole]

Hi Chris, bit late to this party - sorry. Do you think a lidar module would integrate well with this project, a combination of visual and 3d mapping would be quite cool. Plus, on a more robust chassis like the Devastator (?) robot chassis you used on the pi controlled series could be a useful tool for hazardous area searches etc. It would be interesting to see you do a video or two on such a project. And quite educational. Now it's time for me to catch up with the rest of your videos that I've missed. Cheers!

[@dancoman8]

Looks like a smart vacuum cleaner.

[@stillfindingmyway]

Great video Chris, as always! One question, though - during the "table top" demonstration, if you were to place a blockage in front of it, would it still attempt to change direction? I ask this as it seemed you gave it specific examples to work with during that activity.

[@ExplainingComputers]

It might -- I actually never tried! -- as the trained net is still based on an initial model. But it would probably not be as good at avoiding objects, as I did not show it other objects. In the first collision avoidance demo, the robot was running using a pre-trained neural net, where it would have been shown a variety of blocked and free situations, from which it learnt enough to be able to operate in a space it had never seen before. Which is rather amazing if you think about it. It is also possible (although I have not tried it yet) to run multiple programs together -- eg one for collision avoidance (to keep the robot safe, as NVIDIA put it), and another for object tracking.

[@F15HHOOKS]

On a similar note Chris, have you seen the Tesla Autonomy Day presentation on KZbin? The first part details the bespoke chip build used for full self driving that I think you may find interesting.

[@ExplainingComputers]

Again, I will investigate! :)

[@opalprestonshirley1700]

Pretty amazing.

[@Tr3xShad]

Impressive. I was wondering if the nano module can be used on Xavier NX developer kit, as the Nvidia documentation state they share same pin out. Already have the nano module but eyeing the NX development board with it support for M.2 drives support

[@tamaseduard5145]

Thank you sir 🙏 🙏 🙏

[@SlowPCGaming1]

It could make for a neat cat toy. Teach it to perform different stunts or tasks based on which animal it sees under a variety of conditions. Or toddlers, puppies, any small silly creature with curiosity or skittish behavior. Is there a full cover for that bot? I wonder how well it would do on detecting stairs that are carpeted or feature a confusing geometric pattern.

[@ExplainingComputers]

There are lots of JetBot designs now, all for makers to but together and modify. I love the idea of a cat toy! :)