I Made the World's Best Foosball Robot!

My process making Foosbar: the (unverifiably but plausibly) world's best foosball robot! This has probably taken on the order of 500 hours over the last 8 months to make, and you get to see it all in the span of a mere 15 minutes!
Code: github.com/mis...
Twitter/X: / from_scratch_yt
Instagram: / from_scratch_yt
TikTok: / from_scratch_yt
Github: github.com/mis...
Thank you to Teknic, Qualisys and Tornado/Home Billiards for making this possible!
Teknic: teknic.com
Parts: SC4-HUB, IPC-5, PWR-IO-24VDC, CPM-SCSK-2331S-ELNA, CPM-SCSK-2331P-ELNA, CPM-SCSK-2331S-RLNA, CPM-SCSK-2331P-RLNB, CPM-SCSK-2310S-RLNA
Qualisys: qualisys.com
Tornado: tornadofoosbal...
Home Billiards: homebilliards.ca
Music credits:
Rachmaninoff Piano Concerto No. 1, University of Washington Symphony: imslp.org/wiki...

[@Gio81Sa]

*Uploads a video that goes viral *Disappears for 10 months *Comes back with another banger I love this.

[@Turalcar]

The Michael Reeves Stratagem

[@whateverppl1229]

and hes a little kid, 10 months to him is like 5 years. he'd even put Howard Wolowitz to shame

[@Vinz3ntR]

he probably worked 10 months on this

[@blad...]

@@Turalcar That's who I immediately thought of.

[@NoHope-WhatSoEver]

Quality not quantity.

[@NeptuneDesigner]

A new "Stuff Made Here" is born ! Amazing project.

[@impact_42]

If only he could stop using the exact same punch lines as SMH.

[@JohnDoe-f4c2x]

Really!

[@egrabe31]

@@impact_42 Hey, at least he only did it once.

[@dallinmilby8021]

I came looking for this comment.

[@princetikki]

I was about to say this too!

[@renurajamagesh7470]

A god-level foosball robot and it’s remotely controllable? Insane! Quite possibly the coolest toy anyone’s ever had in their bedroom

[@built-from-scratch]

It is until you realize I have 5 cameras menacingly staring at my bed... (albeit not plugged in all the time)

[@BuLLGotcha]

@@built-from-scratchI’m curious if you thought about using some sort of gyroscopic/accelerometer equipped ball instead of the cameras?

[@rider573]

I'd like to see another set of motors added to the other side and he can invite other programmers to challenge his programming. Add in more camera angles and hi-speed replays and you've got something.

[@wadysawkostrzewski8557]

@@rider573 same thing I thought about - instead of robots killing robots in arena - this gem!

[@candyman9635]

@@BuLLGotcha So we need a ball that's the same size and with a similar surface that's hollow enough to contain an accelerometer and all it's gubbins. It must not be bouncy and must be shock resistant. sounds kinda hard

[@johannes.scharf]

With such projects across the internet the "getting there" part is usually pretty boring, so I often just skip to the good part of the presentation of the finished thing. But you've made the problem solving part as interesting as the final demo which made this video overall interesting to watch. Well done!

[@alacer8878]

Bro if this shit doesn't blow up like your last. . . Man, that first seven seconds had me already hooked. It's not enough that you're building a robot, but you gotta pull off feats of skill like *that?* Man, my jaw dropped. Glad I subbed back then. Alsp holy crap the flexing with that blind shot. You're nuts.

[@TravisHi_YT]

The companies that were smart enough to give you the materials for this project easily made their money back with the advertising, great video!

[@isaacnaumenko2053]

Video turned out amazing. Btw That's my hand in the stream at 13:48.

[@built-from-scratch]

Can confirm if anyone had doubts about this prestigious honor, that was indeed him.

[@penapenis]

@@built-from-scratch Did the foosbar ever smack your fingers? XD

[@aze4308]

nice

[@WarFish_Sardini]

Your hand is awesome, thank you so much for gracing us with such beautiful sight.

[@aviko9560]

The hand made the video fr, I'd be otherwise very bored! Gracious sight!

[@simonabunker]

I love that your brother has the confidence of a LLM!

[@mr.dayearly3583]

Was this an LLM? It sounded like he wrote the instructions. Was actually gonna ask about using a model lol

[@braveheart6816]

Absolutely amazing seeing these young creaters spring up all over youtube

[@SamualN]

you could train neural networks by continually pitting them against each other on your real foosball table. you could then eventually (after a long time) and up with a neural network that is very good at irl foosball

[@JscWilson]

Ideally you'd be able to model the game in software so it didn't need to physically play each game for training. It would be interesting to see how well training on a simplistic model translated to playing in the real world. Hopefully well enough to work as pre training, which could be fine tuned with real games. It would speed up the overall training time significantly.

[@salia2897]

@@JscWilson Human players can currently beat the robot cause they have a better intuition of the real physics. So a NN would definitively need to learn this. A physics simulation would need to be very accurate to do the training in order to be better than the hand written software.

[@jessewilson3571]

@@salia2897 "Intuition of the real physics" isn’t the only factor. Foosball is dynamic-another critical factor is how quickly a player or robot can receive input, make decisions, and act upon them. As an extreme example, imagine a robot capable of perfect predictions but taking an hour to process them. Another robot with faster reaction times, but less accurate predictions, could score a goal before the slower robot could respond. More accurate predictions are obviously better (all else being equal), but the robot doesn't need to have a better "Intuition of the real physics" to beat a human. That is why I said it would be interesting to see how realistic the model would need to be to be useful. One approach could involve initial training on a simpler and faster model, followed by fine-tuning on a more realistic yet slower model. Such a strategy might outperform dedicating the entire training time solely to the realistic model - but again, the question is how accurate do the models need to be?

[@salia2897]

@@jessewilson3571 It does not need a better one. But it does need one that is good enough to do the required maneuvers that humans can do. And doing that can only be learned from a physics model that is good enough and that will already be quite complicated.

[@jessewilson3571]

@@salia2897 With a quick Google search, you can find a number of examples of foosball-playing neural networks that have been trained using simulations. This includes a few public GitHub repositories with Unity models that aren't very complex. So, training on simulations can definitely be done with models that aren't extremely complex.

[@ragleysenpai698]

Nice job on this! I designed built one of these as well for Oklahoma State University, on ours I trained a neural network to control the rod actuators. Additionally a core objective of ours was to make it the same footprint as an actual foosball table because the previous version had the side mounted actuators like yours and was extremely large!

[@myrmatta1]

This dude has 2 videos and he's already making higher quality content than 90% of creators

[@jaredf6205]

And many thousands of dollars in sponsors already!

[@JujuProdGames]

The amount of passion into this is unleveled, UNBELIEVABLE JOB :O

[@bmcgee3507]

You're one of my favorite youtubers and you only have two videos! I think it's partially because you have a very similar style to another of my favorites, Stuff Made Here. You made a few (maybe unintentional, but probably not?) references to some of his jokes (only have to do it once, magic wand, etc.) which just makes it better! I can't wait to follow your path to (hopefully) success as an engineering youtuber!

[@dangerkingston]

Dude, first of all this is amazing, you are obviously a true fooser and finally built a foosball simulation worth playing... This is big. I'm so excited. I sent this to all my foos friends... Just the idea that I could zone out and practice solo is a game changer in itself. Tagging on to the end of your video here... I think with tweaks you've already realized that you have the ability to really fine-tune this thing to be unbeatable, But what's the fun in that? Chess against the computer is fun I'm sure, but as with the game of chess, the joy of foosball is really defined by the interactions between two players. I feel like you've might not be seeing the obvious next step: making a whole second control setup on the other side and allowing two people to play online against each other on a physical table. I have an idea here, and I'm going to send you a DM.

[@yolf-n5p]

"Arent you too young to want those servo motors?" "Yes, yes I am"

[@Klos1neMN]

First off, I am in complete awe of what you did, from start to finish the meticulous detail is amazing. Second, this has been shared by many online foos groups so expect some praise and compliments from the foos community from all over. Third, how did you account for the recoil needed to straighten out those rollovers? Seeing those snake shots at that speed gave me nightmares of the one time I played Brandon Munoz and his insane speed. Also, and not to take one iota from this massive achievement, but have you seen the Foos Gadget one? It allows you to record a defense for a certain amount of time and then jump across the table to shoot against the defense you just recorded. It also has presets in a phone app that you can load up and send to the defensive rods with increasing difficulty, it's pretty cool. However, those stick lane passes that your machine does are insane, reminds me of Tony Spredeman lane passes. Lastly, there are some foosers already saying, "He should bring this to Worlds!" (Tornado World Finals, Lexington Kentucky, Labor Day weekend)

[@built-from-scratch]

For the third, I didn't need to do much other than manually tune the timings on the snakes. The rotational motors are insanely fast, so even though the sideways motion to start the snake is faster than the average human's, when the ball is hit it has enough forward velocity that it's reasonably straight (hopefully I'm interpreting what you mean by recoil correctly here). This was actually a bit surprising to me; I originally started with the robot doing a push shot which I thought would be a lot more consistent, but it turns out snakes are much easier to get working. For the first lastly, not 100% which device you mean, from what I can tell Foos Gadgets just sells goal spedometers/automatic scoring. Regardless the goal of this project wasn't to be solely a training aid, I really wanted a fully autonomous table so I didn't spend much time on stuff like a practice app. For the last lastly, unfortunately it's probably not possible. I've already disassembled it (I want my room back!) and I'd like to move on to other projects for now. It's not out of the question though, it would be neat!

[@Klos1neMN]

@@built-from-scratch Wow, I was only HOPING for a response and yet here you are! Yes, you interpreted the recoil I was speaking of correctly - when humans shoot a snake they need to strike the ball while moving in the opposite direction to offset the first rule of motion. From what you are saying, it sounds like your machine does it so fast that it doesn't allow spray, which is insane. It would have been cool to see what it did with a push, I have to recoil my push back to the wall so hard that it practically jars the table. Yeah, I added two "Lastly" paragraphs, sorry. I understand that us foosers probably won't see this at the next World or National Championships but it is impressive and the foos community would love to see it live and get a chance to meet someone who loved the game enough to attempt what you have done. Once again, bravo and keep creating!

[@Astra_Dystopium]

When I was in college I played more foosball than I did studying. This is the coolest thing I've ever seen. You're brilliant my man!

[@renestefanez71]

This channel will grow faster than this robot can shoot the ball. Nice to have i third channel to choose next to Stuff Made Here and Mark Rober

[@Rubynator007]

Super good and interesting video! The longer I watched it the more I was hoping you would try to code it with machine learning in the end, but I guess that’s too complicated right? Just imagining in my head two robots practicing foosball all day and night in my room and eventually become unbeatable. That would be next level.

[@jamenkaye987]

4:29 Dude perfect better watch out!

[@ErazerPT]

Mad props on this, awesome. Only thing I'd try adding, as it doesn't go too deep in the weeds, would be to "take shots from any point you can find a clear path". Even if it misses, it will rebound hard, and your robotic front line has better reaction time than humans so it can chain another shot from there.

[@ryanhacker5259]

I love that you come up with an idea and then follow it through to completion, no matter what it is. This was a treat to watch!

[@passby8070]

wow, pure respect to you for putting everything together in 8 months, you have done way more useful learning than most engineering student who have done 4 years of study. I am sure you already have many job offers, choose wisely in order to put your skills and creativity in good use.

[@jsardi56]

Very impressive. Right up there with the self-solving Rubics cube.

[@FoosballSportsNetwork]

Nice video man. Glad you got a Tornado. That other table would have failed your test 100% of the time.

[@trepidati0n533]

This is a case of where machine learning could really shine. There are so many kinds of networks that could be employed and be trained on. Furthermore, it could train itself since the win condition is pretty obvious. Would be also be interesting to pre-training in a mechanical simulation model before hitting the real model for training refinement.

[@DungeonMetal]

I am a primarily drunken foosball enjoyer, so I don’t encounter it often, but I am an engineer at heart; and holy hell this project is mad impressive.

[@paulsmyers203]

Seems like the logical next step is for Tornado to host robot foosball competitions using their table as the standard playing field.

[@dulesipu]

So glad I subscribed, this is fire. Excited to see what other projects you will work on!

[@alec7446]

I’m sure so many people have thought about this idea. I know I have. Ultimately we all thought.. yeah that’s going to be way too complicated. Really cool that you followed through with it!

[@vildis.]

those motors and cameras are so overkill, i love it 🥰

[@louievenni9287]

The new stuff made here has just arrived to KZbin folks! Love it ❤

[@johnmalin4933]

Awesome work! I'd be really interested to see end to end deep reinforcement learning on this :)

[@khloros17]

This is such a Stuff Made Here idea and I love it.

[@msx80]

Right? It's like a Young Stuff Made Here show

[@peeniewalli]

I like the way it's made, solid table, good motors/controlls. But one thing i like to mention because the 'robot' side playing only 1 human opponent seems unfair. Because it don't need to switch it's "hands" from one handle to next, like the human does. So Little advantage, or play 2 person×robot f.e.. Still, those accentuaters are fast! Program input/output/calcutating is ofcourse superfast. And indeed the camera stuff versus our stereo eye sight. Nice project!

[@jenniferdarline]

Please keep doing this the world needs more people like you. You will go on to do great things for society and the world. remember this

[@arbo6850]

No way, he's back!

[@rev0luzeR]

Dude. Who are you? 😮 This is amazing. KZbin algorithm took me here and i am staying! Keep up the mindblowing work!

[@zinixoficial]

The parts in this project are so overkill it doesn't even feel like an achievement.

[@slakjawnotsayin5451]

I would have put the camera where a player would be watching from, and used tracking as well as prediction to avoid the occlusion problems. I've thought about doing this as well, and also doing it with my dome hockey table. Also, great job actually doing this and doing it as well as you did!

[@Mo_2077]

Can't wait to see what else you make

[@twocsies]

If you set up a robot on each side of the table , then reinforcement learning will be a ticket to unbeatable robots.

[@jonasgraff5767]

This is hands down the most impressive technical project I've ever seen. I've seen the UBC IGEN table, would be hilarious to see this bot destroy another bot in a match

[@treysomerandomnumber]

Would be fun to see the foosball machine playing itself, while learning how to get better

[@viniciusnoyoutube]

You should make a "sleeve" tube for protection on the rods that stick out on the other side.

[@4bSix86f61]

2:50 That cheap servo uses a potentiometer not a hall sensor

[@built-from-scratch]

Okay fine you got me, I added this scene last minute and that was all I happened to have on hand. It's not like the power electronics (an H bridge) or the MCU (teensy) were accurate either though

[@4bSix86f61]

@@built-from-scratch Lol

[@xXIkaron]

Next up: Add motors to the other side, so the table can play against itself, as well as automatic ball drop-in. Then use a neural network to control both sides, and let it train by playing against itself. I bet it'd come up with some insane inhuman strategies. Just have to be careful with local minima, but if you first train it to react like your current manual implementation would (which can fully be done in software at well over 100x real time) it will start out being about as good as your implementation. Then just occasionally switch one side back to your implementation to make sure it doesn't train itself into a weird special case. Leave that running for a few months and see if you can still beat it!

[@PaulSavagecomedy]

“Foosball, also known as Table Soccer” Damn, son, it was right there.

[@nathanpaes9767]

@StuffMadeHere This project seems right up your alley. How about a friendly Robot vs Robot competition?

[@UffDaDan]

Saw about 30 seconds and subscribed like i did for Stuff Made Here. The more the merrier!

[@LondonTablefootball]

Very well done! People in the foosball community are completely amazed

[@JaceTan-90]

I can see this robot being implemented globally where foosball tables are currently located but have no players.

[@EricSundquistKC]

With the right approach, you could train a neural net to take care of all the robotic decisions and end up with a truly terrifying opponent

[@c016smith52]

What an incredible project and video, so glad I found your channel!

[@40watt53]

I've never seen someone good at foosball before so just the first like 7 seconds of this video was mindblowing enough.

[@joefisher3950]

Huge recommendation. Usually if you find yourself coding very small simple tasks humans are naturally familiar with… use a neural network. I’d love to see this in a new iteration but with nerfed action/reaction speeds for human viability and the robot actually being a great player.

[@bando404]

So cool! Now plug another one in and have them play against each other.

[@babywaffles2373]

this guy has uploaded 2 videos and they've both been absolutely insane

[@jasonmorrison7120]

this looks like a perfect application to apply ML/AI to. let it train itself on all of the corner cases you mentioned for fine control.

[@PeteBrubaker]

Paused at 5:19 to comment - for some reason I knew where it was. I'm a graphics software engineer so maybe it was the shadows. But I think it also might have been the angle of all the other players on the board. Anyway, great stuff!

[@Beamer1969]

someone needs to bring all the robots together for a tournament

[@PlanetXtreme]

It is so tempting to build one now but I know I will cry before I even try

[@ammarsalmi]

Protect this kid at all cost. This is the kid that will invent real virtual reality gaming for us!

[@petobuild1922]

I wasn't about to like this but when you mentioned that everything is opensource I instantly liked the video even before you could mention it.

[@interessiertkeinen7227]

I would love to see Tony Spredeman play against this robot at some point. Or maybe you could show your work at the next ITSF Tournament in the US. That'd be sooo awesome!

[@eduardolima5495]

Ohhh boy!! I would love to see all those different robots he showed in the beginning of the video competing against each other!!

[@jackmclane1826]

This scale of this project is astonishing. 500h for a project like this is like lightning fast.

[@randomperson-qj1oq]

i desperately need to see this man doing something with factorio

[@DanielOwens-v5f]

Subscribed! Thank you for sharing your talent, ingenuity and passion with us ☺

[@Maverrick2140]

this would be very much more interesting if it were a competition between teams consisting of people coming up with the computer vision part + AI for maybe creating tactics to circumvent the opponent and ofc the Mechanical side of things (not necessarily the physical construction but also how to use the table ) one system playing against the system of someone else ^^

[@Luca2013333]

That's an awesome project. I myself thought about building something like this 😍 can you try training a Neural network for playing table soccer. I would love to see different tactics that the machine would come up with

[@shottysteve]

i was watching the beginning and i was like... this guy doesnt even have a tornado table... and then u got a tornado table lmaooo

[@drippyer]

8:51 Eric knows the internet well

[@petrsokol588]

Dude, this is so cool!

[@nic23lond23]

I skipped the 10 month wait... now i have to wait for the next one... you are doing awesome sht

[@audas]

Really impressive mate - would love to see passing improved. But what would be great is introducing angles, especially side angles where you slam the ball into the side wall with the players. Considering this - it would be just too hard - so seriously the next logical step - and not too hard would be some machine learning with a bit of AI - tensorflow, OpenCV etc - you may have to model the entire system in 3D with Bullet or Jolt physics library.

[@ILiveInUrWall_]

Crazy how underrated this guy is

[@HenriqueSantAnna]

Great project! I loved to see this video. Now I'm waiting for the AI trained robot version.

[@ryanmcnair3451]

SUCH a clever cookie! You're gonna go far kid

[@dannymars]

I wanna see a tournament with all of the other robot tables.

[@artmalig4228]

Now this... This is engineering. Amazing build!

[@MrBlaDiBla68]

Exceptionally great experimental physics and software here, Cheers !!

[@anonymous_246]

Glad I became part of your YT journey this early.

[@guuzila]

well done, gives me the vibe of micheal reeves combined with stuff made here, keep it up, im glad i got this recommended

[@brianconlogue1302]

2 videos and almost a 100k subscribers.. this channel is going to blow up!!!

[@tsbrownie]

How does it do with pull and push shots? Oh, and back field shots or does it only shoot from the front?

[@MKZ3003]

Oh my gosh this is one of the best algorithm pulls EVER

[@tyruslawhorn]

Brilliant stuff. You have a very bright future ahead of you.

[@jacobokeeffe1071]

If you make a second one and an automatic return mechanism for the ball and this thing would have the most epic endless battles!!

[@omerrivlin2637]

Amazing! And doing all of that in your bedroom... Keep up the awesome work!

[@chriskingston1981]

Would be also so cool to let a neural network just start with random movements and starts learning what to do and figure out how to play this❤❤❤

[@PapaLurts]

One other use I could see is to build two of them. Since you already track the human players position you could just connect two tables and make multiplayer foosball (well online multiplayer)

[@fish1207]

Imagine a table with two of these playing each other

[@matematicaluan]

a sugestion: use your code to make a fly by wire "human controled" by joystick in which the human player just puts the "intention" and the machine calculates the execution, its almost done its a matter of assemble... so you then can use the joystick to play the game even if you dont know that much of... next stage would be regulating the levels of assists in the joystick, just like a F1 game.... brillant videos, please keep going!

[@skanderbelhaj2278]

Dude you’re awesome…. Can you make a video explaining your approach to coding this thing… how exactly you manage to code all the possible cases when the robot handling the ball… how does the robot make an offense strategy or is the same thing done very precisely… i would appreciate it if you do a video about it … thanks in advance

[@lump123]

damn the rach concerto

[@KamramBehzad]

I'm officially gobsmacked. Brilliant.

[@bramweinreder2346]

Great project. Would also be cool if you could program it with different speeds and strategies.

[@fatalmystic]

I wanna see "stuff made here" build one and have a competition :D