Swarm intelligence simulation. Project "Screaming Insects"
ฝัง
- เผยแพร่เมื่อ 4 พ.ค. 2023
- I tried to simulate swarm intelligence.
Blind agents must look for resources and deliver them to the base. Communicate at a short distance with the help of a cry.
Each agent follows a simple algorithm:
Take a step and increase all the counters by one.
If you bump into one of the items, reset the respective counter.
If this is the destination you were willing to reach, turn around 180 degrees and change the destination objective in your head.
Shout out the value of one of your counters plus the maximum distance you can be heard at.
Listen to what the others are shouting.
If you have heard a value less than in your counter, update the respective counter and turn in the direction of the shouting if you need to reach this place.
You can support the channel on Patreon:
/ simulifehub
More support, more opportunities for new projects. - วิทยาศาสตร์และเทคโนโลยี
i've wondered what kind of pressures could exist for a mound-building simulation. maybe it's cold outside and they lose health if they are not near a shelter block
ooo maybe acid rain which they need to protect themselves from with high ground and cover
As far as I can tell, you don't need external selectors. The strange fog was fun but obviously trees in our world don't contend. Similarly, there is no selective pressure on termite mounds other than predators, and even then, the colony always recovers with exceptions. You don't need acid rain.
The rules listed for termite mound creation are largely based on a single fact, no? Well, two. Gravity exists, so an agent can fall to the ground, and if they are touching a wall, their movement is largely unrestricted with the exception that they must touch the wall. Use resource to create wall, then harvest resource, as long as resource is far there shouldn't be bulbous growths of the mound.
The only problem I can imagine is how to convert to a 2d plane. Perhaps agents can move through certain number of walls? Perhaps walls support each other over a channel as long as they aren't more than x units apart, like a vein would collapse but not a capillary so you can have a network of valid paths?
As a fun idea, albeit a challenging one, implementing some actual thermodynamics would make for an unusual and complex challenge.
@@Minty1337 boid algorithms, introduce predators
@@123seven3 If you are talking about the real world, there are a quite large number of selective pressures on termite mounds, wasp nests, bee hives, and ant mounds in our world, with structures having to manage, co2, oxygen, humidity, temperature, food availability, predator defense, and parasite load. The specific requirements change from group to group, but most of these groups have specific groups of workers whose sole job is to maintain the structure’s internal parameters, and specialized life living only within the controlled habitat of these nests.
This is quickly becoming one of my favorite channels - totally fascinating topics
this channel steals videos from another, it leaves no mention of the author, the author foo52ru ТехноШаман
@@kostyagroza3838 to be fair: channel info says "foo52ru in English."
@@kostyagroza3838 and foo52ru channel info says "English version of the channel: www.youtube.com/@wallcraft-video".. so not sure what you mean with "stealing"
@@TheRealBarni11 Oh. I didn't see it
I love this so much. I made a copycat program for myself to mess around with. I added a predator object that tends toward the nearest agent. When the agents shout the distance they think is to the nearest predator, instead of the others turning toward them, they turn exactly away from them if that distance is within a certain threshold. Makes for some interesting behavior-scattering, followed by reorganization.
I'm going to gear mine toward a natural selection simulation. Queens will have genetic information (possibly neural networks) that dictate what they try to do with each timestep (heal, create a certain kind of agent, etc.). Not exactly sure how I want to create the selective pressures for that yet though. I think I'm going to let them make a pseudoqueen that tries to get far away from her and then, when she eventually does, she establishes herself as a new queen. Expensive resource-wise to do so though, so she'll need careful planning.
Excited to see what you do with this going forward!
Can you share the results of your simulation? What you're trying tk do sounds fascinating
@@Deepclow I'm working on resource balancing which is the most difficult part of natural selection simulations in my experience. It's a lot of trial and error, essentially. I have the other core mechanics down; the queens have a decision tree for example that I think I'm ultimately going to replace with a neural network.
I'm on vacation from work next week and hoping to spend at least a full day on the project, so I'll let you know depending on how the resource balancing mission goes!
I also got the urge to make another one that's pheromone-based instead. Oh the endless rabbit hole of "Hey I want to code this!"
Since I see no one trying to explain the termite algorithm, here's my take:
I think one should think about "bar of grains" as pillars of a (locally) cone shaped building.
I also think it is really important for this algorithm to take into account that:
- different agents start at different times and in different locations.
- many termites may be working simultaneously on the same pillar.
Also, when talking about heights, they are always relative to the ones around the specific pillar.
-- Algorithm --
The termites start working on their personal pillars [step 1]
Since some termites started before, they will reach H1 ("height 1") before others [step 1 -> 2]. Those will continue to work on their pillar until they reach H2 ("height 2) [step 2 -> 3].
The others that started later will reach H1 and find out that there is a bigger pillar nearby, thus moving towards it while it has still not reached H2 [step 2 -> 3].
Once the pillar reaches H2, all termites are freed to move to a bigger pillar if it exists [step 3 -> 2], or to create support arcs [step 3].
While creating support arcs, there's still a lower pillar without an arc, go there and build it [step 3 -> 3]. Otherwise go to the "base height" of all pillars and start building another one.
----
So to my understanding, the termites will tend to gather more in the bigger pillars, forming radial structures similar to the forming paths shown in the video.
Thanks. I'll consider your option.
In the video, I used the version from the russian-language Wikipedia. The description begins with: "In his 1979 Jacob Bronowski Memorial Lecture at MIT, Philip Morison gave an example of a termite building algorithm..."
ru.wikipedia.org/wiki/Термитник
@@wallcraft-video Also ants are known to use pheromones to guide other ants, one might implement them as a heatmap. With that it should be possible to also guide and reinforce or degress behaviours.
@@wallcraft-video In case you didn't look for it, here is the full lecture postcivilateum.blogspot.com/2008/10/termites-and-telescopes.html (there is a lot of other unrelated text in the lecture as well, take into account) . It's definitely not described with the precision of a computer algorithm, so a lot is up for interpretation. I suggest you read it if you haven't.
Personally, this doesn't seem like an algorithm that would be good to implement. It wouldn't cause any major emergent behaviour, as it's not an algorithm capable of adaptation and the rules required would be very extensive. Also, this is a very old lecture and current research shows that termite building logic actually uses different more complex rules.
It would probably be more interesting to pursue a different building algorithm, for example something similar to the tree series but trying to optimize structural integrity instead, since that would allow adaptation more.
Anyway, hope you enjoy the source.
@@leonfa259 I personally think pheromones would be implemented as a ghost version of the ants at that point in time, so it will act as the ant would normally, but stuck in place.
I love the concept of this, especially the civilization type idea at the end of the video.
I am starting to wonder whether these ideas are written out and executed with a lot of artificial intelligence on the backend and minimal human interference. I feel like this is what that would look like. Deeply creative work lads.
this has nothing to do with AI tho, however this in addition with AI could like you said give some very creative and powerful thingys
@@gryphonschnitzel7140Isn't deep learning/neural networks just swarm intelligence? They have simple rules and combined with other neurons a global behavior emerges. Something a bit less controversial would obviously be boids that fit the description exactly.
This is objectively the most underrated channel on yt
this channel steals videos from another, it leaves no mention of the author, the author foo52ru ТехноШаман
@@kostyagroza3838 i dont speak that wacky languages so this channel is way better
@kostyagroza3838 if you would have done a little research instead of accusing the creator for stealing their own stuff. you would have found out, that this channel is the english version of the original channel as foo52ru clearly points our on their about page
I find this really interesting. When you think about it, human body is also just cells communicating with each other. A single one of them doesn't have a solution for a problem, but as a whole they do.
You need to combine this with a genetic algorithm!
This is unbelievably fascinating please continue this was a part two about the termite mound building algorithm. Also I would love it if you would post a video with just the algorithms running because they’re so interesting to watch
Amazing logic on not walking straight, and the moving queen and new queens.. mind-blowing, thank u so much for thinking honestly
this brings a whole new meaning to "lightning bugs"
computer simulated organisms and evolution is just so cool to me. I love this type of content!
this channel steals videos from another, it leaves no mention of the author, the author foo52ru ТехноШаман
your videos are mad interesting please keep making them
i dont know how i stumbled onto this video but this was realy interesting to see, verry well explained and quite cool!
screaming agents be like: *AAAAAAAAAAAAAAAAAAAAAAAAAAAAA*
I love your videos, keep it up!
new queens be like: "Im spartacus!" "No im spartacus!"
another great video, I forgot I subscribed to your patreon so I was surprised when I saw the credits lol
Thank you so much for your all effort
This is brilliant!
pure gold here. great job
Huh, I did not get notified for the video..
anywho, great video as always, thank you for making so easily understandable content ^_^
He's back! Let's go!!!!
Да неужели это то, что я думаю?
До чего же дошли речевые технологии!
love these types of videos
very interesting algorithm, i like the simple art style
reminds me of an ant pheromone trail sim i wrote a while back
I like the contrast between how the ant screams in this version use an arcade shooting sound, while the Russian one used eldritch noises
Thank you for this new Video ❣❣❣
I like that you used 3 resources just like most insects specifically ants need. Water, Sugar & protein. Also the concept of lifespans & new queens.
I've got a little bit of experience with programming & after watching this I kinda wanna try my own hand at making a similar simulation.
Amazing video.
All of your videos are great! It makes me motivated to to a project like tihis on my own
This is the best channel for life simulation ever! BTW, can you use the concepts used in the tree evolution to make them evolve "screaming" that is better?
I think about it
Really cool demonstration. Looking to do similar projects in the future.
Tangental thought. I wonder how this can be related to the market.
yessssssss
that would be absolutely amazing
Awesome. Just awesome
That's a good one.
Yes, i also like this one the most
this vid was really fun to watch as i like bugs
So doug doug's twitch chat is a good example of that
It is a swarm of maniacs that work together that try to complete a task together
This was a great showcase of something I learned in a course on Multigrid Methods.
Local updates are great for quickly smoothing out high frequency errors, however take much much longer to deal with the remaining low frequency errors afterwards.
I wonder if doubling the shouting range after some time would be like sampling down to a coarser grid, making the errors more high frequency.
You should add some "resistance" in which the more agents near each other the more likely a few agents will not go with the crowd and look for other sources. You could also have mobility where more resources either slow down or speed up the cell(depending on if you see it as extra weight to carry or as some "knowledge"/"tool"/"strength"). You could also add "crime" where some cells can kill or steal from others. Obviously you can do all the things everyone else does in these types of simulations to make them more complex looking.
Very interesting!
Nice!
I like this.
Reminds me of the work that was being done on the game Limit Theory at one point where the world was going to have an ever evolving economy system of npcs bringing goods around from world to world
Cool. I had just subscribed.
Hey I haven't even watched the video yet, but you had me at "screaming insects"! Wicked cool title!
Greate Stuff! wish there was a simple open-source for this for people to play around with
My observation of swarms is when I am working on digging up a tree stump. Just before the stump is ready to come loose, waves of mosquitoes attack me, as if in defense of the stump. I can tell the stump is ready to move by the swarms intensity. It also happens when I burn the brush pile. I know the fire will stay lit when the mosquitoes form a cloud around my head. The correlation is maddeningly consistent. I am certain there is a causal connection to these events.
I would love to see more screaming incects
Fascinating! (Maybe I need a 'Spock raising a single eyebrow' emoji!)
Can't help but feel this has a most prevalent use in mass organizing and surveillance via mobile device.
Not just insect, this is probably how electrons form a thunder
Air flow and temperature is what guides termite mound design
Anyone else see the similarity to the formation of lightning when the paths are being formed? (around 6:00)
now we need a one where the agents will go to war between queens
Swarm intelligence is really interesting, ants on their own seems just like stupid animals but on a large scale whole colony looks intelligent
For your mound problem, the concept is simple enough. The morphology of mounds are based on the local environment, because ants/termites need:
- a mound that has good heat transport or retention (thickness of the walls, geometry vs sunlight)
- a mound that distributes pheromones properly (just humid and warm enough to let them either stick to surfaces or be carried through the corridors? not sure about this one)
- a mound that serves the function of lung, to let air transport humidity and carbon dioxide out of the mound and import fresh air (porosity of the walls)
- mounds that serve as an environment to cultivate fungal species that they eat (again, right light, temperature and humidity)
Deciding factors in the morphology are therefore:
- Abundance and average height of the flora around (shade, soil humidity, resources, etc)
- Amount of precipitation and wind
- Distance to the equator (how much sunlight they get, how extreme the day/night cycle)
- Soil quality (how much the soil retains water, how decent it is as a building material, how much the soil can serve as a substrate for fungal species, how hot it can get)
- The species of termite/ant involved.
You can take any combination of temperature / humidity / carbondioxide versus fresh air / fungal substrate as "incentives" to create structures in both 2D and 3D, not necessarily in the shape of actual mounds, that fit the role of "base of operations" for the queen and home for the colony. In my version, I'm going to separate 'builder' agents from 'foragers', so that the builders can walk over and deposit structure on any pixel, but foragers have to go around walls and stuff.
The most important excerpts from the paper linked below:
"mechanisms can be roughly divided into two broad categories: the secretion of a templating odor at the nest that undergoes diffusion and advection globally throughout the mound, and the local deposition of a building pheromone in deposited soil pellets that promote further deposition nearby"
"carbon dioxide gradients may be used as templates in nest construction"
"[mounds] where convective currents dominate conduction, can grow in a focused direction due to the airflow transporting the odor in the direction of growth."
“The growth of mounds is dominated by building processes at the surface and limited by molecular diffusion of pheromones through the mound wall."
"As a mound grows, unless more termites are recruited into the building process, the volumetric density of termites decreases over time."
"larger colonies will tend to produce a greater quantity of odor, and this high rate [sic] will result in a large mound radius [sic]."
"the absolute size of termite mounds increases [taller] with the amplitude of temperature oscillations" (of the day/night temperature cycle)
"external wind was found to play a subordinate role relative to the dominant thermal mechanism"
"we have neglected here the effects of active transport of water by termites within the mound and the passive dynamics of evaporative cooling that will change the temperature profiles and thereby the mechanics of odor transport"
www.pnas.org/doi/10.1073/pnas.1818759116
Intresting.
12:10 Original Queen is still alive on the right side.
nice new vid
Nice
Great project! Strange narration 😂
I feel like this could be a useful analogy for particle physics
Great video. Bee hive yourself 😉
The economic simulation idea reminded me of German-style boardgames. What if the unchosen resource piles also increase over time(or add efficiency with more resource per carried grain)? Thus agents also need to choose between close but inefficient resources vs distant but efficient
Such an amazing video. I am new to this so can anyone answer this. Is the explanation given above based on "PSO"?
he made 3 things,
1. ants
2. bees
3. humans in an online envoirement right now
I wonder if there are phase transitions if you add noise! (And what the system on the edge of them looks like.)
huh, this is intresting
the queen can only get one agent at a time, cause they block each other or slow each other down | if there are walls build, then they can move without restrictions and over each other
pog
Once the queen is full health for long enough, she start spitting out building block for the mound?
they are like boids, i think
Probably more ant algos.
Great video! Would love some human narration, if that's at all possible in the future. The voice is a bit weird to me.
This is so cool! What program is this written in?? Like pygame or something.
Does the algorithm assume that each shout includes both the value and which location it relates to? So, an agent would effectively shouting "127 to A" as opposed to just "127"?
In the project, the agent, in addition to the distance, also reports to where.
The pathways look more like the evolution of nerves which is probably the purpose of this video except real life nerves are more random.
Do you have a discord?
Or a subreddit?
I think it would be nice for people who are trying/have implementing these things to share their project with eachother.
they have a patreon and offer discord access for $1
🤔 "what kind of algorithm should be used for ants to build city?".. I think some ideas from "DNA algorithms" can be used. all of them can be described like "if conditions like A than doing B"
Wow, fascinating! I'm curious about which language did you use to implement this simulation?
Processing
interesting
Original video: th-cam.com/video/PTTIZ8EmbEQ/w-d-xo.html
Is this an official English channel?
Yes
I like how you've accidentally created electrons
Ants on their own aren't mindless automata.
This should be used it video games. Like a kingdom or village builders, and the agents are people with job and need for resources.
Something similar was planmned for a space game, Limit Theory. Never happened tho
I'd argue that it should be possible for Agents to consume other agents at chance, whilst agents in general can generate with a random "capacity," increased only by consuming others - to a limit dictated by how many are around. And, well, the queens...
I wonder what happens if the queens vanished?
In the final version, if the counter of steps to the queen is greater than the specified value, then the agent can declare himself the queen with some probability. If you remove this option, everyone will go after the one who last faced the queen
Ребят, официальный канал на русском написал в видео с название About the channel, что дал согласие на это. Кажется это не его канал, но кого то , кому он дал разрешение на это.
ТехноШаман отвечает на этом канале за контент. Сейчас идут переводы старых роликов, потом будут новые ролики и на русском и на английском
4:38 can you record this scene while highligting the actual code that you used to program that checklist that each creature contained
Reminds me of the ant simulators.
Then it s an effect of a distribution of random walks with an action, in interaction...
Just like birds.
I recreated this in processing, one problem I am having is once the agent numbers increase it slows down alot, I tried some optimization methods of my own to fix it, but I want to know how did you tackle that. Also when the agents bump into the food/hive and are supposed to turn around 180 degrees, in my case they seem to spread outwards in random directions, maybe it has to do with the angle they hit the object just before turning around? Any idea about that?
I wonder how the swam intelligence scales with both agent intelligence and the number of agents. Is there a phase transition at some number of agents, where after that they find path (albeit after a long time) but if you only have one agent less they will never make a path?
Perhaps there is a phase transition. But I can't say for sure.
Independent of any conditions, the phase transition agent number 'n' lies somewhere between 1 and ( resource distance / yelling distance ). This n is otherwise affected by:
- obstructions
- size and variability of the rotation
- size and variability of the step
- implementation of the behaviors
Hard to say the exact number, but testable.
What if there is a wall in the direct path between food and queen allowing only to go sideways for a given length ? Will the agent be able to find a path ?
Most likely they can, but for this the wall must be impenetrable to screaming.
Will have to experiment with this.
Thumbnail: when you went to the recent tab
Building feels like it only makes sense if some resources are stationary.
Yes, at least the queen
the video script seems just kinda off, but cool stuff
Hmmm...
What if there are multiple points at equal distances?
Like 6 points beeing placed at the same distance(3 blue and 3 orange)?
They will go to all of them, if they are on same distance, then they don't need to choose, they will not lose anything.
The swarm would likely begin to favour just one of those points more and more, and eventually send no agents to any other points.
@Михайло Шуляцький
@ronjx
Yes, it would make sense, but what if it would act differently?
@@dandabossthesecond3599 don't know about choosing only one, i think it can go to multiple at one moment, but not more than that.
@@mishazerg You can solve all those problems by rotating the points in geometric patterns! Combine that with the economic principles mentioned and you will have the shapes mixed with some random interference from changing lengths and get some interesting patterns
I’m curious how well did this perform, did it require a lot of optimisation?
I tried programming this up in Python, spent a lot of time optimising it to try and get it to run well, but even after a lot of work minimising the run time and writing in quad trees, it kind of hits a hard limit around 800 insects dropping down to about 10 frames a second once the insects group up in lines between food and hive.
What was everyone else’s experience with performance?
I had the same experience with p5js, it slows down alot once every agent starts forwarding shouts, I managed to fix it by spreading out the shouts over multiple frames. It has better perfomance now at 4000 agents than it did before at 1000.
what happens if there are no borders on the edge of the map, would they would all get lost?
I didn't even think about it :)
The scouts will definitely run away