DIY AI weed spraying drone, automatic landing pad/battery swap
ฝัง
- เผยแพร่เมื่อ 8 ส.ค. 2023
- Less Expensive food with fewer chemicals and labor. That's what I'm aiming to deliver with the system you see here. The drone uses AI to spray spray only weeds rather than the whole field as is usually done which means fewer chemicals while maintaining yields. Automatic battery swap, recharge and flight plan updates come from the base station which saves labor.
In this video series I'll show you how I built it in the hopes that this will inspire others to improve on the system. I want to demistify how this stuff works - AI for image classification, precision drone flight and automatic battery swaps are all attainable at reasonable prices and their benefits should reach small holding farmers today so all of us consumers can benefit from less expensive, cleaner food.
I'm hoping the fastest way to make that happen is to make this prototype publicly available so others can learn from it, improve it and move it foward to farmers all over.
Video Series Links:
Drone Build Details - • DIY Autonomous Intelli...
Lander Build Details - • DIY automated drone ba...
- Wiring diagram for raspberry pi and pump
drive.google.com/file/d/1sK5t...
- Take Photos script - use this to snap 1 photo per second as your drone flys around. Then sort photos into folders
drive.google.com/file/d/1eZPV...
- Model customization Script - Once you have sorted photos, update the paths in this script and run it to create tflite, quantized models with class labels for use on raspberry pi
drive.google.com/file/d/1LCtt...
- Sprayer loop - This runs on raspberry pi to identify and spray weeds
drive.google.com/file/d/1Qpci... - แนวปฏิบัติและการใช้ชีวิต
How has this guy been cranking out insane videos with autonomous robotics and have less than 3k subscribers!? The amount of ingenuity it took to pull this off is amazing. You've inspired me to build my own autonomous rover over the coming winter and I'm hoping to have some kind of multipurpose use for it. I want to use OpenCV with a Jetson Nano for object recognition and make it big enough to tow something like a gorilla cart but also have the modularity to hook up an electric push mower, a blade for pushing snow around, a sweeper for picking up grass clippings or other random objects, while also having it charged via solar. Planning to start small and expand from there but seeing content like this makes it seem very possible. Thanks again for the great video.
^^ This is what keeps me going :) Thanks man. Glad you're liking the videos, I enjoy building this stuff
FKN BAYYYEEEEERRRR 😠🖕 fr tho John Deer is gonna either hire this dude or steal his idea and colab with pharma to make a $100k automated weeding system
Check out the SpykerCat project! That's the one I want to build someday lol
Nice idea, @inc0de! I hope to see you post about this next Spring!
I'm doing my part.
SUBSCRIBED!
What is more impressive of the time and effort you put into the video is you sharing everything for others to benefit! I applaud you and your work!
Just found this by random chance, I'm 60+ Aussie, not a man of the land, but a practical person, who happens to know a few people on the land. I know that they don't know how to do this, but I do this know that they have a big enough network of people that they know, that they would be able to do this. Thanks Nathan, I've heard generalised information about the technology, but you have explained this beautifully. Will be giving them the link to this video, and don't worry, I'll be investigating wot else you have.
Excellent! Let me know how your friends receive it. I'm hoping to connect with others!
I was working in uav precision ag research in the late ‘90s. Big clunky 2m fixed wing aircraft and big clunky multispectral imagers on board with slow post processing to determine target areas. How things have changed! Great job illustrating the concepts and integrating the newer tech. 👍
This is Awesome, the economics of this thing only improves with weed recognition and camera quality (res) I love it. 👍❤️
One quick thing you can add to improve the battery swapping is to have a seperate "anti spark" pad ahead of the positive (or negative) terminal. This should then be routed through a resistor to the PDB, thus you first charge up all the capacitors on the escs, becs etc before connecting the low resistance terminals. This way your copper tape wont be eaten away by sparks each time you connect
This is a brilliant idea. I'm already working on a v2.0 and I'll include this. Thanks so much for the suggestion!
Very nice. Hope that many engeniers will improve and than sell this design for a low price to farmers
What I like about the idea of using drones in farming, is the potential side effect of allowing smaller individual fields to be workable and economically viable, allowing for a return to greater diversity in the landscape, with hedges and variety of crops being grown side by side, maximising wildlife corridors, curbing disease spread, allowing a large, economically viable commercial farm to look like a patchwork of smallholdings from the past.
Exactly! John Deere has been buying up companies and looks poised to introduce image classification/selective spraying but it'll cost a lot. This get small holding farmers in with the same advantages for themselves and consumers.
This is really incredible work. Deserves way more recognition!
What an awesome project! Keep up the amazing work, and I hope others can see the potential and help develop systems like this.
Respect Nathan. Simply Genius
I salute you Nathan !
This project is so well thought out and presented, really clever stuff, thanks Nathan.
Continuous defeating the weeds should also save on chemicals. The smaller the weeds are, the less chemicals you need. And you can even use physical killing those plants without chemicals. I expect that this has huge potential.
Sweet vid man. I'm in controls myself. Very cool.
Dude! This is a marvelous system. Your willingness to share your hard work is laudable.
Great work dude
Crazy amazing work with things around the house! I'm working starting up my metal work and machine shop and you are giving me a lot of inspiration!
I love it man. Thanks for sharing this with us.
great job man!
Two words: pure genius. Thank you on behalf of the many.
Love it. Nice man.
Amazing! Thanks for sharing this info.
many thanks for this informative video, i like how you make it simple describing AI application in crop protection and weed control, i hope you can spot more on improving the AI model and make it more precise for weed detection
Saw this on hackaday, what an amazing project! Subscribed and liked.
Wow, absolutely genius! Great job we need more with your aptitude and ingenuity! Good job, you got a subscriber in me.
Brilliant! Even though I not a farmer, i'm sharing this video with everyone I know, this is attainable and you're a genius 🖖
I'm impressed by how the camera output is usable given the downwash.
You've earned this subscriber.
Your work and mine are very much aligned. Hopefully I'll get back to working on my videos again soon.
Super inspiring! Thanks for this❤
This is an impressive build! You’ve done a remarkable job and I learned a lot that I’ll try out on my future projects. Nice work!
That being said, it’s not hard to see why a commercial product would be much more expensive. They need to pay developers, support folks, and generally create more finished product and all that does add up.
Incredible!
Just found you. Amazing videos. Definitely subscribed!!
Thanks for sharing! I hope to attempt this project this winter. I have 20 acres to spray.
Let me know how it goes! 2 improvements I'm working on
1. Get rid of threaded rods, instead make it more like a garage door - think bike chains turned by the motors pull in the collection arms. It'd be faster and more reliable
2. Add more batteries, build around them so you don't have to solder the pads. I'm working on a 3d printed drone base with a 3d printed battery bay and 3d printed battery holder. Probably will get to it this winter.
This is incredible thank you 🙏
You're bringing some great things to life. I want to talk to you about another great project for farming....
Next up:
Cute little forklift droids that go and ferry the drone to an open charging station.
Making them cute enough will be a challenge. The community will need to come together.
Good job!
Great job!
I'd be keen to see a video on the change to your field. A before and after if you will.
Perfect !
Amazing
I am proud to admit that I accept the download from your brain into my........thank you Sir.
Man this is crazy
Absolutely amazing work. Echo the comments of the other dude wondering how you have so few subs.
A lot of complexity comes from the drone landing/inaccuracy/battery limitations. I wonder if a cablecam type system might be simpler in some ways. Harder to deploy over a large area like a field but for gardens/yards, I could see it solving some problems. Could take a nice bite out of pesticide use in lawns. I don't have the ingenuity to pull it off but you certainly do.
That's an interesting idea! I'll give it some thought
Patent this and create a marketing plan. This idea is worth money and if you don't someone else will. Then no more cheap tech
I'm actually thinking of building an autonomous drone crawler / tank one day that can selectively cut down weeds on our steep meadows. Your approach looks very promising and might come in handy whenever I get to it.
Looking to do this too! Any luck?
Blackberries have rights too!
Awesome video! Just popped into me feed and now I'm subscribed. Think the next step would be scaling up the motors and speed controllers. Make it so it covers a lot more ground.
Yes, exactly. I saw an 8ft square ag drone in my neighbors field. It was totally human operated, but man it could carry a lot of liquid and cover a ton of ground. They did 60 acres in a few hours.
I hope you patented this, even if you want to open source it, you need to make sure it stays open source.
And I gotta say, your principles are awesome dude.
LOL... Patent. Hilarious.
I don't know that you could get a patent of much value here TBH. There's so much prior art on the underlying tech (image classification, precision drone flight, automated battery swap) because it's been described and used many places. Putting it all together seems like it doesn't clear the non-obvious bar.
At the least though, I've put this out in the public domain which should prevent some other entity from patenting this exact idea. My intent is NOT to patent this design and to keep it open source so anyone can pick up and improve on it. I should look into how to actually open source license the design. I'm familiar with how that's done on software, but not something like this. Ideas?
Great project! I want to build my own but I bet the only expensive thing on the drone is that 'orange box' 🤷. (Edit) You could use the Makita style battery connector, its 2 pin terminal is available in most markets. Just need to print the holder.
Very interesting project Nathan! So did you choose the mobilenetV2 and use transfer learning, keeping original weights for the feature extraction? or did you retrain the model completely after adding your classification layers? Thanks for sharing!!
Sir, this is your kindness and generosity to share the details of your work with us for free. I am a university lecturer in the area of genetics and plant breeding. My hobby is building and flying FPV drones and quiet experts in that. However, due to a lack of knowledge in computer programming, I was not able to add such an AI to my autonomous drone. I will follow your instructions soon and will inform you if I am successful. Once again, I truly appreciate your kind sharing. This is very serious to realize that agriculture cannot be done in the old ways anymore and we need to adapt new ideas and technology to be able to fight climate changing and produce food for ever growing human population. 🙏🙏🙏
hi pedro, what do you think about all gubments drawing a line around you(Antarctic Treaty) and saying you are not allowed to leave? If you don't know what I mean, read my about tab.
This is great to hear! Let me know how it goes, I was hoping to inspire others to try this and improve on it. Honestly, I was surprised how far image classification has come, it really doesn't much to train your own custom model.
When you train your model get LOTS of training images - hundreds at least, thousands are preferable. The accuracy gets better and better.
That landing platform is a perfect example of less is more. I wonder what it would take to protect it from inclement weather?
Linseed Oil
@@WW5RM or some shellac
Polycarbonate sheet or similar for the landing pad a covered aircraft hanger for the battery swap area🤔
I'm hoping next version of this will simplify the landing pad even more. I'll swap out the threaded rods for bike chain so it looks/acts more like a garage door opener. I also want to add a lot more battery bays while simplifying the swap process - still a work in progress.
Came from Hackaday
Kudos. Subb'd. Watch your back - Bayer's stockholders won't be the only ones pissed.
My favorite mad scientist, another amazing video. Do you have any plans to scale this further (outside of youtube of course)?
Perhaps... I get the biggest kick out of building it and seeing it work, I don't have much drive to commercialize. That said, I'm hoping it'll inspire others to improve on it and a future goal would be to get like minds into a community to actually push it forward (open source design or something like that)
A similar setup could be very helpful fighting wildfires
I am really not trying to make a thing but I am interested to know how this fits in the US regulatory system for drone spraying. Specifically FAA Part 137 and state pesticide rules. Since it's so small it doesn't need the over 55lb exemption which is amazing and is now the current bottleneck for AGRAS drones which are all now over 55lb. Also, this could be really amazing for some of the work that we currently do. This is closer to a flying "wiper" as they are referred to in AG for the small individual plant pesticide applications normally carried around by hand or backpack.
It still requires FAA certification - I cover that at the end of the detailed drone build video. It's very lightweight as you said so the process is much simpler than for >55lb drone, but the pilot license/lightweight exemption is still required sadly. I hope it gets simpler at some point.
Seems like these would fly really low, hopefully an exemption could be obtained for agricultural use at low levels some day. This could dramatically reduce runoff into waterways with widespread adoption.
What about the regulatory side of the chemicals? Where I live, there are little to no herbicides approved for any aerial applications, and if they are, it's with significantly more water and coarser spray nozzles. I think that will be a bigger hurdle to mainstream adoption of drone sprayibg than anything else.
I’m looking at AgOpenGPS (open source auto steer system) that can run relays to shut off sections on my sprayer to reduce sprayer overlap. Now I’d love to have something like your tech running a see-and-spray program. I imagine it would be easier (probably famous last words there) to teach the model to recognize what needs to be sprayed during a spring or fall burndown where the only green things in the fields are weeds to be removed.
Thanks for sharing all this valuable information, this will probably be a winter project for me.
Thanks for turning me on to agopengps. I hadn't seen it before and it looks great!
Yes - turning on/off sprayer nozzles would be a perfect application. The time consuming part is getting LOTS of training photos and classifying them correctly. The more photos, the better, think thousands. John Deere and a few others are doing stuff like this but they have lots of people who can view/classify. You could also look for fungus/evidence of pests so you can spray fungicide/pesticide on the right places. Putting it out so others can add to your repository and make the model stronger is a great idea too, perhaps on agopengps.
Let me know how things go! This is just what I was hoping for.
thanks for this , it's what i'm looking for. I've built an rc push mower, and I'm also wanting to put a sprayer and a device to spread tick crystals during summer using ArduRover and start planning missions.
also what if you could have the drone fly over and identify the weeds, then have the gps lawn mower pull a spray tank and then the mower will do all the spraying?
That sounds like a fun project! My fields are pretty tall so the mower would look more like a tractor! Next year...
Consider not just farmers, but also certain condos could benefit.
The farmer can do the plant image capture on foot so that it can be more accurate.
Just what I need to build to combat Kudzu here on my property.
I think it is still a requirement by the civil aviation authorities in some countries for the drone operation to be supervised. The technology is still not yet able to be fully autonomous. It’s flight can still be disrupted by strong winds or malfunction. When the RTK fails, the drone can still go haywire
Great job, you could streamline it with cheaper parts and sell it as a kit, but still you have to be an engineer with crops to do this.
The engineer/farming overlap is often higher than people realize, especially on smaller farms. Most farmers have enough knowledge to repair their own equipment, and pick up new info pretty quickly. There's a reason Right to Repair is having it's best effect in farming
Hey, i am very impressed! I love your solution for the landing platform and for the battery swapping. How about the AI-part, do you already have some experience with reliability and hit rate?
Yes, the hit rate is quite good for me, but I'm lucky - the grass (off target) and blackberries (on target) look so different that the model doesn't have much work to do, like >80% accuracy. I had about a thousand training images.
To expand this to handle more diverse crops and weeds will take a LOT of work and images which I haven't begun to tackle yet. I'm hoping to get more people involved to submit their own training images for that :)
Now that it's been a few months, is there anything you'd change if you were starting over from the beginning? I've got a 2 acre hobby vineyard and just got my part 107 license and would love to build something like this. I've been talking to GPT4 about it for hours but it hasn't come up with anything near this level of ingenuity. Hats off to you, sir.
Glad you enjoyed it! I'm working on version 2.0 now and I've got lots of changes.
- rather than a threaded rod to move the "collecting arms" run them with a chain, similar to how a garage door is moved. The motors I used are actually used in garage doors as well I found out later.
- More battery bay spots to charge more batteries
- 3d printing most things to make it simple to build. I want to make the drone landing gear one big hollow pipe, the battery cage a 3d printed box (so I don't have to hand solder so many wires and can just use the connectors on the battery itself) and change the battery bay into a rotating wheel (think like a Ferris wheel) on each side, so when you push one battery out of the bay it also pushes out the spent battery into an open bay on the other side of the drone and it just starts charging there.
I'm planning another video with all those changes added in in the summer. Still working on the 3d printer designs at the moment. Let me know what you create!
Vegetables are 100% chemical. Chemistry is the study of matter. Legit UAV work though, cool!
Great idea! I had no idea RTK was this inexpensive now. If you can do this with a Pi 0, imagine what you could do with a Jetson Nano. At 15 watts for the smaller ones, it does much more AI and still takes less power than the motors. A better camera (more light/higher shutter speed) would also be better for humans fine-tuning a model and the AI running it. Both may require a larger/heavier drone.
Yeah, a bigger drone + better compute power is the next step. Do you know how rpi + coral tpu compares to jetson nano for performance and weight?
There's a new global shutter camera for rpi that has to be in the next version, will be way better images
@@nathanbuildsdiy I looked up the jetson nano for a different project using whisper (OpenAI's free speech to text/translation) which needs 16GB of RAM. With your own carrier board it would be very light, like a Pi compute module, but that's more work than their off-the-shelf boards. The software stack looks flexible with pytorch or tensorflow. I haven't bought one yet.
I only briefly skimmed the Coral system. It's probably less power hungry, much less expensive, and less powerful. I haven't used TensorFlow Lite, but it's designed for edge applications (like phones not servers). pytorch seems to be winning for research; many projects switch from tensorflow to pytorch, but I haven't heard of anyone going the other way. Both packages are quite capable, and TensorFlow Lite is probably quite power efficient since it uses 8 bit data. I deal mostly with Large Language Models like Llama-2, and with these 8-bit is more than enough once the model's trained.
So the Jetson Nano is more $$$, maybe slightly heavier & more power hungry, but way more RAM and better compatibility with random github projects. The Coral is probably more efficient, cheaper and lighter especially with the Jetson's full heatsink. I'm not familiar with image recognition libraries other than OpenCV, which could be overkill. You've done more image training than I have. Suppose you could also connect an FPV camera to a desktop computer and run whatever AI system you want. Good luck!
Cool project, I wondered if you gave any thought to running multiple drones simultainiously?
Not yet, but certainly for the future. Audupilot can actually drive swarms (I think, haven't tried), so I'm hoping it wouldn't be painful to add.
would be nice to see field before and some time after if this actually works or wind from drone makes it miss
I'm adding this to my mower
LOL
Great video!! How many acres can the drone cover before needing a battery swap? How much liquid can your spray tank hold and can the drone handle a bigger one? Thanks again!!
I can get about .5 acres in a run - both the battery and liquid. Some places weeds are really dense though and it just sprays and sprays and runs out too quickly. At least it's automated so it'll just keep refilling and I can hit that area again later.
I feel like we should be friends
Ha! I just checked your channel and realized I've been watching your videos. I love the spinning drone with wings.
Yes, we definitely should be friends. I'm an EE not aeroE, so I've got plenty to learn about the drones, but I'm fairly good with the electronics and code parts.
👍
A 10K resistor between the Pi pin 2 and 3 and GND should keep the floating pins low at boot and not interfere with their signaling while the drone is in action. Or a 10K resistor between EnB and GND on the L298 would work also.
Excellent advice. Will do for the next version.
Perhaps add a non-toxic fluorescent dye to the herbicide and then fly over at night with a UV light and take video (GPS coordinates included in video) to confirm where the spray was laid down
Very cool idea! I'm also going to grab before/after photos once I have FAA approval to spray
Not to mention it requires a special FAA permit/waiver to release chemicals from a drone. 14 CFR Part 137. Buying/making a drone and using it to spray chemicals from it = a serious no no.
Correct. I addressed this in the detailed build videos. At least it's
We can better stabilize that RPI camera. The rolling shutter doesn't help either, but it should be stabilized so it's not as blurry (also reduce jello effect)
I think there's a new camera for rpi that doesn't have a rolling shutter, that would be a huge benefit. I also think I need to start 3d printing - the camera holder, the battery housing, the landing gear/tank for the next version. That would likely help stabilize the image, right?
On your battery swap platform, it would probably be quicker to move the sweep boards with a motor pulling cables back and forth instead of a threaded rod.
Totally agree, after building this I thought... why not make it work like a garage door with cable/bike chain? Way simpler and faster, that'll be in the next revision.
Also could you use the camera to aim the landing at a symbol on the landing pad? Might be a way to get your landing location more precise
This is pretty cool, doesn't it need to be lower to spray the weeds without wind scattering
This does ok right now since the downwash of the drone motors really sends the spray down. It coudl be set to any height with the lidar. I have a few tall things sticking up in the field though, so if i go much below 2m I could accidentally run into those since i don't do object detection in the forward direction. Something that could be added pretty simply though.
Typically pesticides are sprayed from planes which fly much higher than 2 ft off the ground.
This is amazing. Speaking of open source, could you help some like-minded freedom-(as in free speech)-loving Ukrainians with their particular needs? There are many small drone builders who could use some help. I can put you in touch.
You could probably use a combination of an electromagnet and electrical contacts to do the battery swapping (think XT60-like with magnets, or the USB-C ones for your phone charger), rather than modifying the battery with those big copper contacts (and having the corosponding ones on the drone).
Totally. I'm thinking next version will have a 3d printed battery case with normal connectors to plug into the battery inside the case, then some big plugs on the outside so it can plug into the battery case holder (also 3d printed) on the drone/charging station.
This is really cool. I'm using your work as a guide to clean solar panels with a drone. I am curious, how do you determine when to fill up the water tank? Has it ever been empty during a mission before landing on the landing pad? Thank you.
Yeah, it has emptied. I'm working on a larger capacity tank and I try to keep missions short, but there are areas that are totally covered in weeds so it sprays constantly. A float to detect when things are drained would be a good future addition as well
@@nathanbuildsdiy Thanks for the response, I thought along those lines of using a float sensor to check the water level. It's a bit tricky because I want to keep the drone as light as possible but I will investigate. Thanks again!
Second suggestion: raspi is having low shutter speed, resulting in those blurry photos. Consider using either the newer model RasPi camera, an action camera, or a cheaper DSLR that you can plug in and connect to the RasPi.
I guess it’ll be too heavy
Yes, I just saw raspi has a global shutter camera which (I think) means it exposes everythign at once rather than scanning the image. That'd be just what you're talking about and wouldn't add weight.
I was wondering if task separation might make sense. One drone used for target acquisition and then another one or more for delivery.
Yes, that's a good idea - fly high the first time to decide where to go, then zoom in the second time.
@@nathanbuildsdiy That too, but I was thinking of having two drones at least. One would be responsible for tagging locations of all the weeds to target and the other(s) would carry the payload directly to the targets.
Hello, I want to build a more simpler Drone, it's my first one, Something that can use some liquids and some ''grains/powder'', it's for my father that works on crops in Brazil, some of the stuff is kinda impossible to find here, just a drone that can be ''kinda auto'' after config or can operate at home/notebook or radio can work out with a FPV kit and after trying to make it work with the pump? Plz help! ty for the content
Far out!
Isn’t it RPi 0 W version 1 ?
Is there a way to use a commercial drone with a 4K camera and image stabilization to do initial reconnaissance and classification with an RTK antenna for accuracy and deploy a second drone just for spraying? Or is it simply too complex?
Yes, and I think raspberry pi now has a global shutter camera (lower res, but will remove the swirly photos caused by drone downwash). That sounds like an idea for future revision
Maybe you should consider Lithium Ion battery rather then Lithium Polymer. Li-ion has almost double capacity at the same weight. Keep in mind though that the discharge rate will be less. However this drone flies rather slowly and probably doesn't consume to many amps. Make sure you get the high discharge 21700 or 18650 cells
This is a great idea and something I didn't know about. That'd be a huge weight savings, I'm going to add it to the plans for the next version. Thanks man!
I would just have changed the raspberry pi to a dedicated neural processing unit for neural network, because a raspberry pi will take more than 10s and a npu such as coral dev board will take about a 1 sec and cost about 80 for the entry board
I'm thinking a coral TPU add on perhaps for the next one. Thoughts?
Has the UAV ever landed facing the wrong way? Or is this just not an issue?
No, but only because I pull a "trick" when it lands. It follows the same 3 waypoints at the end of every run so it's always moving south to north. That orients it facing north when it lands since the drone always faces "forward" when it flies. That's how I can then push it into the corner in a known orientation to swap the battery
Use IBT-4 instead of L298N H-bridge
IBT-4 weights less, and holds biggest current
Great suggestion. I had l298N laying around so I used, but really not ideal. Glad you know of a better one.
Cant see how the drone path and spay accuracy is good if you need a massive landing pad. Great idea though
Wind. The GPS is accurate to a few cm with RTK base station corrections, but some wind while landing can push it off right as it comes down :(
If the setup is right next to a field, how many acres per hour can a unit like this cover?
I'm severly limited by # of batteries and charging time (the drone sits idle while battery charges). It can only do ~5 acres per day. I think it'd be more like 25 acres/day with proper batteries. BUT I've seen a few ag drones in the area that are 8 feet across (all flown by people, manually controlling)! They fly much faster, carry more pesticide and can cover tons of area. That'd be the really valuable scale up. I saw one cover a 60 acre field in an afternoon
Won't spraying weedyside this way affect the crop around it?
Yes, but currently the whole field is sprayed (crop duster for example) so this cuts WAY down on how much is sprayed since it's targeted.
Cool tech, great build, misguided thought process on how weeds weeds should be handled. I know it's not your personal train of logic, but it's still sad to hear it perpetuated.
what's the bitrate of the pi? could you put some jet fuel on the drone by using a Jetson nano?
It's pretty slow, like 3 frames/second. A coral TPU or jetson nano would make a huge difference.
@@nathanbuildsdiy For 3 Frames a second, it works quite well. I think the one thing that's stopping farmers now is the "just works" factor. That's the hard part :)