Evolving Genetic Neural Network Optimizes Poly Bridge Problems

AI is kinda limited to its code, it doesn't change it

I have an exam tomorrow about genaric algorithms and i saw this video. I hope it will help me.

Wait, thats all? I thought you´ll use that to actually play the game. Thats bummer but oh well.

You should look into Frida, it's a Python package that allows to inject hooks into executables for specific calls or values. Basically CheatEngine on steroids with a scriptable API. You could interact with PolyBridge's engine directly (ie. for fitness calculation) instead of having to reproduce it.
Not an easy tool to master but you may save a few days by spending a many months learning to use it!

As a student who learned about Genetic Algorithyms it always amazes me how hard it is to define reward function so the "AI" completes the task. Sometimes It takes more human inteligence to tell AI what to do than the intelligence the AI can reach

cant play the game but can waste too much hours and days to make an ai to beat the game

shout out to the people who got the architect joke

Very cool but I really do think AI is the scam of the decade. Imagine if some actual company claimed using AI to build an actual bridge; Knowing it's just this waste of time and energy exercise of automatic trial and error, I'd much rather pay twice as much to engineers who'd reason about decisions they make rather than just tweaking knobs on a simulation program

Istg these ais taking virtual magic engineer's jobs

89.9k sub!

$100.000!!!???

BTW, BRIDGE LIKE THAT IS NOT ORIGINAL

TRUSS BRIDGE DESIGN IS ALREADY EXIST IN 1820

I used to play that old freely available bridge builder game at school as a kid. My solutions looked like the AI thumbnail but hey, they worked!

"Genetic algorithm always cheat."

The most valuable lesson learnt in this video

day before this video turns 1 year old 🎉🎉🥳🥳

Maybe double the mass of the ball, forcing the system to find a way to stabilize it. Or maybe do recreate a golf cart, where if the cart tips over, it's game over.

bro I discovered your Chanel and I love it I hope you get back posting bc of your editing style that makes it more interesting and fun . Without your game changing video I would have know about this . I hope u give us an update

Go "Crazy Machines" =D

Why are you using a custom "Elapse" function instead of "FixedUpdate"?

Ok.

makes me wonder if AI will ever be able to actually have its own intelligence, as the way reinforced learning works is defined completely by the reward function made by a human. And currently it is more time efficient for the human to solve the issue themself, rather than constantly tuning the reward function to give the desired outcome. It's just not possible to translate a human problem into a reward function for an algothrim.

Ah, another good creator I find after they leave... How bloody great

Why don't, you all, make mods that implement ai into original games? Why are you making copies of the games?

Ok so im extremely late to this video but why remake the game? Couldnt you use melon mod loader or something like that so you can get the data you want directly from poly bridge ?

When I build an AI, I like to take my progress, save it, change a variable, and continue

It's easier to tweak when the results are nearly instant

I used your AI model to help my son build a popsicle stick bridge for a class project. He got an F.
UNSUBSCRIBED.

i wonder how much would change if you added more rules to how the score was calculated.

for example: would the AI make more bridge-like structures if you penalised them for having a wobbly bridge (the nodes of the bridge having a high velocity during the simulation)

food for thought

Video was kinda underwhelming, you spent way too little time looking at how the ai solves problems and what crazy solutions it comes up with. I disliked the video, but i appreciate the work you put into making this, coding is not easy

"I'm too busy to play this game, so I recreated the game for an AI to play it."

RCE reference real

I have absolute no clue how ai Training works but wouldnt it maybe be more effektive to give the ai the soluten of the lv. So it knows how a bridge should look like and then start mutating from there, by rewarding the least amount of pieces/weight used.

Have a fitness test that measures the 'wobbliness' of the structure so that the model tends towards a static object.

Another point of reinforcement should have been the speed of the "car". The faster the car reaches the flag, the better BUT:
if the velocity of the car is faster then a set value (overspeed) or doesn't move below a certain threshold: penalties.
there could be another penalty, if the car drives backwards. just some ideas i had.

I love to play unity games

When the video ended I said out loud "damn it's over?"

I’m way too busy to play the game so I recreated it (also playing the game to test if I do everything correctly) and use a genetic algorithm to solve the problem.

Sounds good 😂

Would recommend a better initial population so that good parental features are preserved. Also, need to add dynamic constraints. I have a document explaining a similar project in my git repo.

Minimise the cost of material, please, that was great

what if you gave a neural network 30 billion neurons?

Ai needs to die, let us do the creative things

Was that a RCE reference??

Polybridge 3 civilization goes hard

So, you didn't do poly bridge. You did your own game.

AI ultimately finds a cheat. Seems very concerning.

That was so cool, kinda wish it was much longer haha

Could you not have used FixedUpdate and set the tick in settings

muy corto el videooo

To be fair, i truss more the human-built bridges