Do you mean how we detect the green turning signals?

@@evbrobocup yes

We employ a recursive algorithm to find "groups of green pixels" and then use the average position of the black lines around it to determine its relative position to surrounding black lines. This information is then used to determine the final turning direction. Feel free to check out the TDP on our website if you want to dig deeper.

Ah too bad, sorry to hear that. But if it makes you feel any better, we totally screwed up Superteam last year too (0 points in both runs...). This year's challenge was a lot easier though, at least with a working line follower and another team with decent rescue capabilities

​@@evbrobocup Since we were 1st in the standings we got put together with a team in the bottom quarter, so we spent most of the time rebuilding their robot and got it working well enough towards the 2nd run. The challenge at the WC was much harder though and required us to communicate multiple things including a variable. Surprisingly enough communicating from a EV3 brick to a Raspberry Pi using Bluetooth is very reliable.

Ah okay that sounds similar to our challenge at WCs last year. We also had to communicate with an Ev3 team, but Bluetooth didn't work at all for us (there were too many signals interfering with each other, outside the venue it kinda worked). But we did use a Pi 4, maybe the Pi 5 has better Bluetooth hardware? Or was there something you did differently that we did?

​@@evbrobocup I think my reply got deleted or something, maybe because it had a link? Anyways in short the Pi 4B/5 both have BL 5.0. The Ev3 only has BL 2.1, so its limited by the Ev3. We had no issues with range, interference or walls. There are examples in the GitHub repository (including all our code) I'm gonna publish on our GitHub organization in around 1-2 weeks. I might/probably will make a tutorial on how to get it working properly, since the only guide on the internet about it is flawed and the example code is broken.

@@evbrobocup The Pi 4B/5 have the same Bluetooth version while the Ev3 has 2.1. We're gonna release the code in 1-2 weeks, it needs a little work to be readable.

That's certainly a really good question, as we faced the exact same problem a few years ago. You can actually see in some of our older videos (mainly the ones from team Proxima) that our motors would just randomly stop sometimes or the robot would get stuck on bumpers because the slight increase in needed torque caused the motor to stop completely. There are some possible solutions to this problem (note that it is mostly caused by physics though, because motors need a lot of power to start spinning but with low PWM they don't get a lot of that): 1. Try to drive faster. That may sound stupid, but often your maximum speed during linefollowing is caused by a single bottleneck in your hardware/software design. We still use the very same motors from years ago, but we don't have problems with them anymore because we can drive a lot faster nowadays. 2. More efficient H-Bridge. We have absolutely zero experience with this solution, but our motor drivers (L293D) are really inefficient and a better one might give you the slight boost you need. 3. A software hack: Every time your motor stops moving (are you using encoders?) or drops below a certain PWM value, you can give it a tiny speed boost at higher PWM to get it going again (eg 100ms at 255 PWM). I wouldn't recommend this option for the long term though... 4. Get motors with lower RPM (higher gear ratio). Depending on what motors you are using, there might exist a drop in replacement with the exact same dimensions that just uses a different gearbox and therefore provides lower RPM out of the box (we use the Pololu 20D motors with 140RPM and there are several other versions of our motor). Out of all the options I personally think that option 1 and 4 are the best ones. Maybe you can try to improve your linefollowing first and see if your algorithm can handle higher speeds. If that's the case, you don't even need to buy new hardware. Otherwise I would just try to get my hands on another motor with lower RPM and higher torque.

​@@evbrobocup Do your robot's motors always drive above a minimum speed to prevent this?

No, our linefollowing is just fast enough so the PWM signals the motors receive are naturally high enough to not stop the motors

@@evbrobocup we got pololu's 20d motors instead of the 25d we were using and they're working a lot better than the 25d at low speeds, thanks for the advice!

No problem, glad I could help

Despite silver detection, our linefollowing is purely algorithmic and essentially calculates the average line angle on the current frame which is then used to determine the motor speeds. The angle is calculated by iterating through each black pixel in the frame and using basic trigonometry to find out its angle to the bottom center of the frame. We then sum up all angles and divide by the number of black pixels to get the average. The actual algorithm is slightly more sophisticated (for example we give pixels custom weights based on their horizontal position), so feel free to check out our TDP for further information as well as pseudo code.

Maybe our motors are just "slightly" overpowered with a max power of 4x27W = 100W

Unfortunately our neural network for detecting the victims is quite slow, but this is most likely a software issue as other teams with similar hardware manage to detect the victims with comparable accuracy in only a few milliseconds. There are also some hardware acceleration boards specifically designed to make the Raspberry Pi more suitable for AI applications, but we did not have the resources to get into that nor did we want to tackle our software issue with expensive hardware.

We use the Pi Camera V2.1, but if you plan on building a similar robot any camera with sufficient speed (30+ FPS) and low latency should be fine.

After finding a group of green pixels, we cut out a quadratic region of interest that is slightly larger than the green turning signal. This ROI now contains the green dot itself and a few black pixels that belong to the black line of the intersection. To decide which way we have to turn, we calculate the average position of the black pixels in this ROI. If the position is in the bottom of our ROI, we should ignore the green because there is a black line right below it. If the average black position is in the top right, we turn left (because there are black lines at the right and top of the ROI) and if it is left, turn right respectively. I would recommend checking out section 4b of our TDP (can be found on our website under "BitFlip" -> "robots" -> "2024") for further information.

I guess it will take a few more years before the robots are taking over...

We use a Raspberry Pi and an Arduino. The Pi does all computationally expensive tasks such as processing the current frame from the camera while the Nano does everything else like driving motors and reading out distance sensors. They communicate with each other via USB

Yes, we use four 12V DC motors from Pololu

@@evbrobocup I see. I think I will use it too.

While you are right that we use a camera, we don't utilize a neural network for linefollowing (we do for detecting victims though). Our algorithm is actually quite simple and mainly involves calculating the average line angle using basic trigonometry. If you want to learn more about our approach please check out our website

Or bad code 😅

The claw was manufactured using a 3d printer, you can find the design files in our GitHub repo (link the video description). Silver is recognized using a simple tensorflow lite neural network. If you have further questions after checking out our website and repo, feel free to write us an email.

We don't have tutorials (yet?), but feel free to check out our website and GitHub repo (both linked in the video description) for more information about this project. Otherwise I would recommend the OpenCV line following tutorials from the KZbin channel "Out of the bots"

We write all of our code on our main PC and if we want to test it on the robot, we upload it to the Raspberry Pi using the SFTP protocol over a hotspot from our phone (of course you can also connect the Pi to your WiFi). That way, we don't have to write our code on the slow Pi, but rather on our main machines which is a lot more convenient. After uploading we then use ninja/cmake to compile the code and afterwards run it on the Pi. We are using the standard Raspberry Pi OS on a Pi 4B. In idle the Pi should consume less than 5W and even during intensive computer vision tasks it should not exceed 10W. For initially setting up your Pi you should be able to use your phone charger as long as it can deliver at least 2A at 5V. Just make sure your brick isn't too fancy as they screwed up the hardware design of the first Pi 4 version so it can't handle so-called "eMarked" cables. For usage in the robot, we use 4x 18650 cells in series and a proper step down module that can deliver stable 5V (You can find the exact module we are using on our website). If you have power problems the culprit is usually a cheap step down module that has a lot of output ripple or just can't deliver as much current as they tell you on Amazon. And as for your last question: we all finished German highschool and are already attending university. Unfortunately that means that we will turn 20 this year and therefore won't be able to participate in RoboCup Junior anymore...

@@evbrobocup Dude, thanks for the tip, I just didn't understand this part of the operating system, as if it were the recommended system. But there are so many that I can't even count them on my fingers hahaha Regarding the robo cup junior, it's sad but that's life, I'm in my second year of high school and I participate in a competition called OBR (Brazilian Robotics Olympics). and I wanted to bring this world of computer vision to my state and in the future to my country, Brazil doesn't pay as much attention to technology or robotics. This is annoying because it's what moves the world today. You must have already won the robocup at least once, so I have no intention of using your code because it would be cheating with the other competitors, I want to develop a whole code. But I'm sometimes in doubt about what to use on my robot and how to do some parts, so I'll ask whenever I have doubts.

Honestly it shouldn't matter too much which operating system you use, but I would still recommend Raspberry Pi OS as it has the longest track record and proven to be reliable. I wish you the best of luck, Computer Vision and robotics are indeed the future and also quite fun to mess around with. If there are any further questions just drop us a comment or an email

Thanks a lot, even though this video is quite old ;) In the shown video we did not actively search for the victims but rather drove around the whole rescue area to collect as many balls as possible. But if you are interested in victim detection feel free to check out our newest videos. See you there!

@@evbrobocup thank you very much! I'll watch your newer video.

Unmöglich war der Parkour sicherlich nicht, aber ich persönlich finde auch, dass hier etwas übertrieben wurde. Der Roboter passte ja nicht einmal zwischen der roten Ecke und Säule durch... Bezüglich Wissensweitergabe geben wir unser Bestes, u.a. durch unsere Website auf der sich die meisten Infos über unsere Roboter (nach etwas Gesuche) finden lassen. Mal schauen, ob wir in Zukunft auch noch leichter zugänglichere Formate ins Leben rufen. Dir auch alles Gute, war nett dich kennenzulernen!

@@evbrobocup Danke. Ja, war mir auch eine Freude.

Hi, if you are new into robotics or RoboCup in general I would recommend to build your robot using Lego as it allows for rapid prototyping. However, we've built this robot using a custom 3d printed chassis and custom PCB we ordered from china.

It was extremely bright at the venue which our neural network wasn't used to. But we made some adjustments and were able to get it working again at run three

No, unfortunately we only qualified for the European Championship in Hanover

Of course, feel free to check out our GitHub repository in the video description. Good luck tinkering!

Do you mean this RoboCup season? It will end with the World Championship in Eindhoven this year (mid July)

btw i want to know where to get that omniwheel like that size and what is the best for both pwm control and communicationing with raspberrypi 4b.

cuz i tried to use ardunio mega to control some of motors from signal of raspberry pi 4b. but i couldnt (the signal from raspi 4b is on and off when i did it.

The Omni wheels are some kind of off the shelf solution, so you should be able to find a similar product by googling it. Unfortunately, we have no idea anymore where we bought ours... For controlling motors you can either use an Arduino or the Raspberry Pi, both should work. We usually go with using a dedicated microcontroller like an Arduino for motor control because the Pi has no hardware PWM which makes motor movements inaccurate and prone to mistakes

tysm for replying <3 im sorry for my lack of words. i want to pwm control 4 or more motors with raspi. so i used arduino also used together raspi 4b for more pwm ports. but arduino can't process the signal from raspi 4b in real time. therefore i want to know what microcomputer board is best for processing the signal from raspi and having 4~ pwm port.

What exactly do you mean by "can not process"? Which protocol are you using? I can recommend SPI, I2C or USB/UART communication which all should work perfectly fine using the Arduino + Raspberry Pi combination. In fact we have used all options listed above and can confirm these results.

Are you into RoboCup? It's a robotics competition for pupils, so don't expect something mindboggling :) But I can ensure you that the students are trying their best to improve even further

Of course, here you go: RoboCup Junior is an educational initiative that aims to introduce students to robotics and artificial intelligence through hands-on competitions. In the Rescue Line sub-category, teams design and program autonomous robots to navigate a simulated disaster zone, following a line on the ground while avoiding obstacles and rescuing objects that represent victims (we take part in Line as you can see from the videos). You can find official rules when you enter "RoboCup Rescue Line official rules 2024" into a search engine of your choice.

I have already answered your second comment, but in general it's the robots goal to follow the line and rescue the silver and black balls into the colored corners.

Sorry for taking so long, but it took me a while to find the part number of our wheels. You are absolutely right, we use old Lego wheels as you can probably tell from our videos. YoutTube is quite strict when it comes to posting links, but when you enter "lego 32019 86652" into a search engine of your choice you should be able to find them. Also feel free to check out our website or GitHub Repo (link in the video description), where you can find high resolution images of our robot.

Props for noticing such minor details :) TL;DR: EvB was the name of the school we graduated from, but now we act independently. To give some context: Team BitFlip went to the "EvB Gymnasium" and created this KZbin channel to upload their runs (we called ourselves Freewin 2.0 and πthon πraten back then...). After we finished high school last year, we still wanted to participate in RoboCup, but chose another qualification tournament than the teams of our school. That's why we also registered under "Fantasiename" instead of our schools name. We initially used this channel for solely uploading our runs, but of course some other teams of our school wanted to publish their videos here as well. We (or mainly me, Lukas) do all the maintenance related stuff like answering your comments! Edit: As team BitFlip will officially retire this year, there will be more and more videos from other teams of our school and fewer of team BitFlip. We will see if we make a second channel for that (managed by actual pupils) or keep uploading on here to make use of our reasonably large following. Maybe there will even be some other cool projects we showcase on here, but we haven't really put much thought into that.

​@@evbrobocup Thanks for answering. I've already wondered about it when I was at the German Open. And yeah, I'm a big fan of you, so of course I pay attention to the details.Thanks for answering. I've already wondered about it when I was at the German Open. And yeah, I'm a big fan of you, so of course I pay attention to the details. 😉 I'm really looking forward to seeing more teams of your school, especially since I would be glad to also see some teams that are actually at your own level so you can relate to the videos. Because most of the videos you can find on KZbin are videos of like teams in top 5 or maybe top 10 of a country's Nationals. And, I mean, my team's pretty decent too, but well... not that good hahaha. However, it's sad that you'll have to retire from RoboCup Junior this year. I would also really like the idea of seeing some other projects from you guys, and I will definitely keep following your projects or whatever you'll decide to do. You're kind of my role model team. I wish you all the best, especially for the European Championships, regarding that will be your last championship at RoboCup Junior before retiring🍀

Thanks a lot for your kind words, I'm really glad that we seem to have a positive impact on someone with our videos. I am sure that you will become even more successful in the following years. Just take a look at our old videos from a few years ago and compare them with our current robot... You will certainly progress equally fast if you keep improving your robots :) If you have questions feel free to check out our website or write us an email!

I guess you mean BitFlip's robot :) This one is from team Krypton

As I have written in the other comment answer, the video shows a very niche competition and is not meant to get millions of views (nor being extraordinarily high quality...). And I am sure that the robot will improve in the future :)

I am truly sorry that you did not like our video. To give you some context:The video was recorded on a very crowded competition venue and is not meant to get millions of views. There are just some RoboCup teams that upload their runs to KZbin, so other teams can rewatch and learn from them. Somehow KZbin thinks that random "non-RoboCup" people could like these kind of niche videos... Anyways, thanks for watching and taking time to comment!

@@evbrobocup No apology needed, nothing hurtful done! I really admire the tenacity being involved with robotics requires and thank you for capturing and sharing this moment. Glad this can be watched and reflected upon

Lol 😂

We use standard Lego wheels for the front axis (we resalvaged them from an old Lego set) and Omni wheels for the back

We wouldn't have been able to fix the problem if it wasn't for kind teams that let us use their 3d printers :) But thank you anyways!

Cool to hear that you are getting into RoboCup! We use a neural network with the MobileNetV2 architecture to detect the balls. Past experience taught us that purely algorithmic solutions (like OpenCVs HoughCircles function) don't work reliably unless used with advanced image filtering techniques.

@@evbrobocup Thank you for answering my question! I will refer to it and proceed with the development!

We only have two motors which control the front wheels. The back wheels have no motors so we cannot control them.

Thanks for your interest, in Germany we have seven qualification tournaments, where you need to qualify for the German Open (usually the first few teams from each of the seven tournaments are allowed to go to the German Open). At the German Open the first team can participate in the World Cup and teams placed 2th to around 5th are allowed to participate in the European Championship. So we have some kind of hierarchy that goes like this: Qualification -> German Open -> European -> World Hope this helps :)

Team BitFlip is qualified, team Krypton (the robot in the video) unfortunately not. Really looking forward to meeting you in person! Lukas from team BitFlip

​@@evbrobocupwe will meet there, good luck for everything

You too, looking forward to meeting you!

@@evbrobocup ☺️

Yeah of course: [email protected]

Thank you soo much! But also props to team Overengineering^2 and Offroad who were 1st and 2nd this year

Thank you, of course you can ask. As I can't paste images into my answer and it would become quite long trying to describe exactly what we are doing, I would recommend checking out our TDP if you are really interested: kraemer123.de/bitflip/files/BitFlip%20TDP%202024.pdf If KZbin removes the link in the future, you can also find the TDP in the docs folder of this year's GitHub Repo or on your website (you can find both in the video description)

@@evbrobocup Ohh, thank you very much for the answer, it can help me a lot!

No problem and good luck! If you have any further questions, just drop us an email

In this video we were indeed using Opencv for linefollowing ("Out of the bots" has an awesome tutorial series on this topic), but we did not actually search for the balls. What we instead did was to drive row after row through the rescue area and whenever we encountered a wall, we closed the arm and checked the arm position using the built in encoders. If the position differed from the actual "home position" when the arms were completely closed, we knew there must be a victim in them.