Видео

Love this

Too big brained for me. Piggie!

I preferred the previous video with commentary, but this is also cool. Commenting for the algorithm to be honest.

I was wondering why there was few carnivores until you mentioned size. It is always beneficial to be a big herbivore, just look at Elephants, or the Sauropods Also, I just found this, me likey

I hope you're doing new episodes Or trying recreate first ones

this has to be a game

It broke the game😂

What's the soundtrack used for stage 4 called?

i think maybe having your territory moved is advantageous in some scenarios

i think the speedier predators came from the higher amount of yellow area

yes, if the simulation continues, the species would enter an oscillating pattern, where there are die-offs and resurgences, till they can find a cyclical balance.

Ah, yes, the new bear-themed horror concept: Pentakill at Polar Bear's

You should add another stat called intelligence and basically the higher the stat the better choices they make, if to fight another one for food or teaming to jump another animal.

feels wrong that rainforests have low precipitation

i think you forgot to make video where you move from grid based sims to these xD. also the one doesnt have tile features etc

i think it is kinda mistake to not set reward for staying alive IF there is moving danger on map. it could at least try dodge as well . instead you got crazy bombsquad that just run towards target in hopes to get a lucky run for 'Closeness' to the target

I don't remember you talking about sexual reproduction, but you mention finding a mate here so clearly you are doing sexual reproduction in the simulation. So I'm wondering how you "recombine" the parameters of each of the parents to produce the children? I'm also wondering about how you select mates, is it similarity of parameters? I also don't remember if you talked about populations degrading the environment? I'm wondering if you generate the image of the map for each generation? It seems like you could save a lot of processing time by skipping that except when you actually want to see it. I'm imagining that clustering algorithms would find a lot of use with what you're doing. You might want to do clustering on the entire population to find out how many different species there are. You clearly are doing something like that already and I'm wondering exactly how you decide whether something is the same species or a different species? But I can see clustering being used more extensively than that. For example, you could cluster species by environment to see if there are certain kinds of species that inhabit certain environments, and other kinds that inhabit other environments. Maybe you already do this. I think the videos don't show nearly the depth of what you actually have going on. Principal component analysis might allow you to discover relationships that are not obvious.

I can't figure out what you're using the word "build" to mean. Perhaps species?

I'm hoping that this video was a lot more heavily edited for how many runs it covers than it claims. It feels like the panda makes it across once and then the whole thing gets taken to the next level, rather than being left at that level to improve, so there are still runs at the 380K mark where the run appears to end because the panda fell flat on its face right at the start. Also, the way that the panda just walks into things rather than trying to go around them suggests that it isn't capable of telling the difference between safe and dangerous obstacles. It's a good concept, and I'd love to try it myself (I've been meaning to try some evolution programming for a while); I just feel that this tried moving too fast (trying to get it to pathfind before it could walk properly) and in the wrong direction (trying to get it to pathfind at all when it didn't appear to have the sensors or recognition ability to learn about obstacles). Maybe add more explanation about what it can actually sense and recognise for context.

Great info. I'm struck with the fact that you don't actually have a genome associated with creatures. Instead, they mutate by direct mutation of their phenotypes. It avoids the problem of how do you map between a genotype and a phenotype. And yet one of the most interesting things about life is that it does that and that the space of mapping between the two is so big that pretty much anything can be done. Evolution can solve practically any problem.

I love your simulations! I'm hoping they will. "stimulate" me to do my own simulations. I can't help thinking about the complexity of all the relationships and how it's kind of arbitrary what you choose to introduce in the way of complexity. And the real world has an infinite amount of complexity to it, even to the part of the real world that doesn't include any organisms. I keep thinking that it's important to find questions that need answering to explore with the simulation. But when questions depend on lots of complex variables that could be set up in innumerable ways, it's questionable whether a simulation can be helpful. Somehow, it seems like it ought to be possible to go back to some sort of "first principles" when setting up the simulation, striving to make the results of the simulation be as generalizable as possible, regardless of what the actual specifics of the environment and its complexities might be. Maybe at some point an idea that's actually useful will occur to me and I'll be able to give you a suggestion you might be able to do something with.

I'd like to see an attempt to explain results mathematically. Can you find equations describing aspects of the simulation, that have stable points that correspond to stable states of the population? When you build in more stealth if the creature eats more plants, or vice versa with animal eaters, it seems to me you were building in rules that you actually want to see evolve. It will be interesting to come up with specific questions and then see if a simulation and a course wanting mathematical model can answer them. I've been accumulating ideas about this for a long time but just haven't got around to trying to sort through them and try to put together a simulation to address an interesting one. Regarding sudden mass extinctions, it would be interesting to study each one and find out the cause of it, which is one thing you can do from a simulation, to see if you can come up with any factors that seem to often lead to mass extinctions.

FINALLYYYY

Great video, great simulation! If I understood it right, you have a set of fixed traits that can be present in greater or lesser amount which changes with mutations. I know that one important facet of organism phenotypes is that immune systems and protection against prey are extremely important. They evolve the most rapidly of any of the traits of any organism. And I don't think you had a trait or traits that represented the ability to defend against predation. I'm sure you're gonna be adding that and I'll be watching your subsequent videos.

Something I think you should have in your model is gyro for the different limbs. So it learns posture and or the ability to crawl properly. Unless you have that.. Then I'm not sure, but it feels like it must be missing something else in its input parameters. [edit] are you passing the distance to the danger objects as an input? Like it senses getting closer to a danger object vs a wall or even open space? I'd love to see how you work through what the inputs and outputs should be.