Head to brilliant.org/NeatAI/ to get started for free with Brilliant s interactive lessons. The first 200 people will also get 20% off an annual membership.
I already did the weight variations in a version of boids i've been working on, altho the result is more like a state machine with templates than a sliding scale. They are attempting to emulate an ant colony... but in space. With glowbugs :D
Could do something like a three actor simulation: - prey (sheep) eat grass but fall asleep eventually, making them exceptionally easy prey - predators (wolves) eat sheep and slowly starve if they don't get any - shepherds shepherd the sheep from the pen pasture to a safe sleeping spot, but they can't shepherd at night
Are you passing any information into the network about where the final target location for the boids is? Otherwise the drone will just chase the boids arbitrarily until the target location is reached.
Yes, it knows how far away the centre of the flock is from the target. But that distance is the length of the path through the maze they need to travel. I use it as a component of the fitness function
I think a more realistic representation would be to give the AI the position information of all sheep rather than the center of mass. Since the boid algorithm uses the CM as a parameter, giving this to the AI would be adding information that wouldn't be known otherwise. Forcing it to calculate CM on its own would create a much more robust solution! Still great video, I especially loved the flow fields
But then it would be difficult to deal with varying number of sheep. What do you do with the input when you add 10 sheep? Center of mass certainly doesn't provide complete information but it helps a lot to simplify the problem
@@sebastianjost True it doesnt simplify it, but it does make it more general. maybe since its a generic algorithm making the neural nets, at some point, theres 2 inputs for each sheep. or maybe the algorithm finds something we cannot think of too!
Thanks for the feedback.. ! As its boids, they will tend to stay together (same as sheep), so went with the centre of mass.. Once I give the sheep their own brains and a bias to escape, I'll have to send each of their positions to the drone(s)..
@@sebastianjost use the transformer-style topology! It handles a variable number of inputs well by repeating the same modules over them, then automatically paying more attention to the most important ones. The original transformers use fully-connected key, query, and value layers, but I wonder if NEAT could be used to evolve those modules
@@user-me7hx8zf9y you're right, it is. But since it's not very difficult to understand (i.m.o.) and solves very similar problems, I think it's reasonable. Of course it wouldn't fit the channel name very well.
@@neatai6702 that is the best approach. Keep doing what you love and I suspect you will find enough people on this slinking oblate spheroid who are also interested in such things, that you will do just fine. Honesty, your natural self will carry the day, tenfold over. I look forward to your future videos. Happy software engineering! FF x
Head to brilliant.org/NeatAI/ to get started for free with Brilliant s interactive lessons. The first 200 people will also get 20% off an annual membership.
Wonderful ! I have to go back and look at NEAT again.
that's pretty NEAT
This is so cool drones for everything!
O my god, you got a sponsorship, I remember when you had 400 subs. So proud of you, great channel
Yes! Thank you! Really enjoyed my first year on youtube.. 30 videos, 17k subs and a sponsorship..
I already did the weight variations in a version of boids i've been working on, altho the result is more like a state machine with templates than a sliding scale. They are attempting to emulate an ant colony... but in space. With glowbugs :D
I want to see the laser drones fight each other....
I’m gonna tell my cattle-dog about this!!!
Thanks Dom !
Another great video! Thank you for making it and sharing.
Thanks for the feedback !
Good to see your sub count going up. Well deserved.
Much appreciated
As always - great vid! Can't wait for the lasers :D
Please do a course.
Could do something like a three actor simulation:
- prey (sheep) eat grass but fall asleep eventually, making them exceptionally easy prey
- predators (wolves) eat sheep and slowly starve if they don't get any
- shepherds shepherd the sheep from the pen pasture to a safe sleeping spot, but they can't shepherd at night
Yea, I'm working on that type of pred prey setup to see what they evolves..
cute idea!
Thank you! 😊
Amazing Content as always 👌
Glad you think so!
Neural net trained drone? More like "Neat AI's videos always own!" 👍
How much faster is a quad tree vs a fixed grid? Did it only help for sparse populations?
Are you passing any information into the network about where the final target location for the boids is? Otherwise the drone will just chase the boids arbitrarily until the target location is reached.
in a cul de sac situation this would be horrible i think, but information about where the herd should go next may be useful
Yes, it knows how far away the centre of the flock is from the target. But that distance is the length of the path through the maze they need to travel. I use it as a component of the fitness function
Why python? Do you compile it, instead of running it as slower interpreted language? Just curious
huh i thought it was as in physical drones
Merry holidays and healthy new year! :)
Same to you and yours !
hello .Neat!Bravo! that is outstanding editing- farewell~;))
Thank you! Cheers!
HOLY SHEEP THJIS IS AMAZIN\''''
Thanks for the feedback !
I think a more realistic representation would be to give the AI the position information of all sheep rather than the center of mass. Since the boid algorithm uses the CM as a parameter, giving this to the AI would be adding information that wouldn't be known otherwise. Forcing it to calculate CM on its own would create a much more robust solution!
Still great video, I especially loved the flow fields
But then it would be difficult to deal with varying number of sheep. What do you do with the input when you add 10 sheep?
Center of mass certainly doesn't provide complete information but it helps a lot to simplify the problem
@@sebastianjost True it doesnt simplify it, but it does make it more general. maybe since its a generic algorithm making the neural nets, at some point, theres 2 inputs for each sheep. or maybe the algorithm finds something we cannot think of too!
Thanks for the feedback.. ! As its boids, they will tend to stay together (same as sheep), so went with the centre of mass.. Once I give the sheep their own brains and a bias to escape, I'll have to send each of their positions to the drone(s)..
@@neatai6702 YES! I cant wait to see what you come up with! I love your videos and I love AI
@@sebastianjost use the transformer-style topology! It handles a variable number of inputs well by repeating the same modules over them, then automatically paying more attention to the most important ones. The original transformers use fully-connected key, query, and value layers, but I wonder if NEAT could be used to evolve those modules
I solved this problem way faster by just getting a dog
Is this inspired by the episode in Clarkson's Farm
I'm sure it played a part.. Great series..
Interesting video!
Though I'd love to see some other AI techniques other than NEAT. Q-learning might be an interesting next step.
Q-learning is a large step...
@@user-me7hx8zf9y you're right, it is. But since it's not very difficult to understand (i.m.o.) and solves very similar problems, I think it's reasonable.
Of course it wouldn't fit the channel name very well.
@@user-me7hx8zf9y Enforced subpopulations and evolution strategies, in the other hand, aren't that far off though
Code please!!
Analogous to cyberspace algorithms shepherding the masses.
Still waiting on your Neat AI trading bot implementation on python
Please do this soooo 😃
Gathering some live trading data with it at the moment..
0th comment! Also, your videos are amazing. Looking forward to seeing you get 100k subs and beyond!
Thanks for the feedback.. I'll just keep plugging away and see what happens..
@@neatai6702 that is the best approach. Keep doing what you love and I suspect you will find enough people on this slinking oblate spheroid who are also interested in such things, that you will do just fine. Honesty, your natural self will carry the day, tenfold over.
I look forward to your future videos. Happy software engineering! FF x
11:26 Brillant Ads ..dislike
the educational and inspirational value is great, but you have no sense of aesthetics
at least it doesn't get shut down by youtube gods for copyrigths :p
@@martinferrand4711 Even squares and circles with well chosen colors can be prettier than some random free graphics mashed together.
I always seem to run out of time for the aesthetic elements.. Something to work on in 2022
So the project failed and after all that work it could not heard the sheep. Why not just show us the ai trying anyway?
did you see the video at the end where it herded the sheep? did we watch the same video? are you intoxicated?