YO great stuff. Cant wait for the next vid! A few tips, maybe punish them for hitting the walls/spinning. Could also give them a broader field of view.
I don't think you explained what the brain output does, but looking at your code it looks like one neuron does rotation and the other acceleration. This explains why they evolve to spin. It's easier. The rotation neuron is always some non-zero value so the agent keeps spinning and the only decision to make is when to accelerate and the answer is when you point in the right direction and there are no obstacles. Involving the other neuron in a decision which way to turn at each moment is more work for this artificial brain. It brings the joke about rubbing two brain-cells together ;)
It would be interesting if you explained how the neural network's output relates to the turtle's behaviour. By that, I mean: what does the neural network actually produce? Simple context-free moves (i.e. a definite sequence of blind moves regardless of the obstacles)? Instructions based on what the turtle detects in its vicinity (e.g., "if there exists a solid 40 pixels on the right, turn left")?
Thank you for the comment! You're right, I forgot to discuss the network outputs in the video. Each network produces two output values: one is used as the turning force, and the other is used as the thrust force - so the outputs are directly fed into the physics system for each AI. Since the network calculates the output based on the input values (the sight lines and the compass), the AI can have "reactive" behavior based on the obstacles around it.
The spinning confused me at first. When you suggested that it was due to the three sight lines, it clicked for me and I immediately thought of the 'run and tumble' movement seen in some types of bacteria (there's a wikipedia page for "run and tumble movement" if you're curious).
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
![The moment we stopped understanding AI [AlexNet]](/img/tr.png)
How does this have 591 views. Most underrated channel I've ever seen.
A follow-up video would be NEAT!
I also WANNt it!
PLEAAAAAAAASEE don't stop making these videos, they are sooooo entertaining and educational at the same time. Thank you so much for making this bro.
Agreed 👌👍
+1
I don't know why this channel is so underrated. The quality of the content is too good
These are really high quality videos for concept and demonstration. Keep going consistently and you're going to be a very popular channel
you deserve so many more subs man.
i rarely comment on videos but you have a great way of explaining stuff.
Doing my PhD and this helped me
Amazing work, man. Keep doing this content ❤
YO great stuff. Cant wait for the next vid!
A few tips, maybe punish them for hitting the walls/spinning. Could also give them a broader field of view.
I absolutely love it man keep it up. ♥
Only 83 likes? Amazing video! When can we expect a video about NEAT?
Congratulations on the work, I loved the project.
Thank you so much for giving out the code
this is too underrated
Amazing video, glad i found this channel; instant subscribe. , Love the video style and the explanations.
I can't wait for the video on the neat algorithm!
Great video! Thank you very much for sharing it. This was an instant subscribe, please keep it up! Also, your video production is amazing.
Thank you, I’m glad you liked the video!
Yup, this was super fun. Thanks argonaut! Looking forward to the NEAT one!
Great! hope you will get back to making videos some day.
This channel is amazing!!!!🤩
I don't think you explained what the brain output does, but looking at your code it looks like one neuron does rotation and the other acceleration. This explains why they evolve to spin. It's easier.
The rotation neuron is always some non-zero value so the agent keeps spinning and the only decision to make is when to accelerate and the answer is when you point in the right direction and there are no obstacles.
Involving the other neuron in a decision which way to turn at each moment is more work for this artificial brain.
It brings the joke about rubbing two brain-cells together ;)
I also like the music.
Bravo!
Bro your presentation is amazing. What do you use? I want to learn! Make the NEAT video please!
NEAT video! :)
Great...just great. Gracias!
Neuroevolution is going to be gigantic in the future. We just need a lot more compute to do the really cool real-world things.
It would be interesting if you explained how the neural network's output relates to the turtle's behaviour. By that, I mean: what does the neural network actually produce? Simple context-free moves (i.e. a definite sequence of blind moves regardless of the obstacles)? Instructions based on what the turtle detects in its vicinity (e.g., "if there exists a solid 40 pixels on the right, turn left")?
Thank you for the comment! You're right, I forgot to discuss the network outputs in the video. Each network produces two output values: one is used as the turning force, and the other is used as the thrust force - so the outputs are directly fed into the physics system for each AI. Since the network calculates the output based on the input values (the sight lines and the compass), the AI can have "reactive" behavior based on the obstacles around it.
@@argonautcode Cheers for the clarification!
The spinning confused me at first. When you suggested that it was due to the three sight lines, it clicked for me and I immediately thought of the 'run and tumble' movement seen in some types of bacteria (there's a wikipedia page for "run and tumble movement" if you're curious).
How does the internal compass work? Does it feed a normalised angle difference, or the cosine/dot product value between compass and direction?
Awesome work! Could you please share the tools used to create the animation?
liked + comment + subscribed
I liked everything from explanation to visualisation.
I have one question, do you write the code in python, and how these visualisation were created?
I used C# / Unity for this video. You can find a link to the source code in the description if you’re interested!
you should give them eyes to see certain distance far so it may stop the spinning
I find your videos to be extremely interesting and wanted to ask, what kind of software or coding would you use to make these simulations?
cool
Will you include the gor your demos?
So so sick
Will you be making more videos?
I’ve been a bit busy for the past few months but yes - I have a lot of ideas planned!
When are you goin to make the vid? I just wanted to know...
Maybe they are spinning to look around them
*Play one more second of the video :* 7:02
was NEAT video ever published?
can you make the codes abd visualizations ooen source like 3b1b, and you you did such an amazing animations
How did you make this?
Yo, where's the NEAT?? Also, please give code for Unreal. Thanks!
Where is NEAT?
How can I get in touch with you?
Vision made of 3 pixels is quite enough of reason to continuously spin
No! burglar!! This idea fell out of my pocket and you found it!!😭😭😭.
🥲