shamichvolga Posted 5 hours ago Share Posted 5 hours ago Hi everyone, I want to show a project and ask for help with the part I can't figure out. What it is GenePool is a Darwinian physics sandbox that runs in the browser. Every creature is an articulated swimmer: its body segments, fins, colours, muscle rhythm and a 30-input neural brain are encoded in genes. There is no tech tree and no upgrade menu. Nobody selects the winners. A creature has to physically swim through water resistance, find food, judge potential mates by their traits, court them, and survive long enough to pass the genes on. That sounds like a small difference from other "evolution games", but it changes everything about how it feels to watch. You are not optimising a build. You are watching a population argue with physics. What actually evolves - Hunger thresholds and mate preferences (both were hardcoded rules before; now they are genes). - Diet, body proportions, fin placement, muscle rhythm. - Armour, burst speed, and autotomy — prey can shed a limb to escape a predator's jaws. - Predator hunting behaviour, including something that looks like pack hunting. Nobody scripted teamwork; it appears when it pays off. On top of that there are ecological systems: key-lock pathogens that spread through dense populations and punish monocultures, water currents, a map editor for painting food/meat/heat/cold, and a petri-dish mode with climate zones. A whole world can be saved to a file and reloaded. There is also a headless mode for long runs without graphics. Numbers, for the technically curious: 112 body morphology genes + 57 ecology genes per creature, a 30-input neural network, TypeScript, WebGL, multithreaded simulation. English and Russian interface. Try it (no install): https://shamichvolga.github.io/genepool/ Source: https://github.com/shamichvolga/genepool The open problem Here is the honest part, and the reason I am posting rather than just dropping a link. Over long runs — 50k to 100k steps — almost every pool converges to the same solution: one small, fast, flexible worm. It is a very good generalist, and it outcompetes everything specialised. Biodiversity graphs plateau. Predators are worse: they either eat their prey to extinction and starve, or they disappear within the first few thousand steps. Things I have tried, with honest results: - Temperature gradients and climate zones: partial effect, now off by default because the cost outweighed the benefit. - Food specialisation niches and food size: some effect, not enough. - Water currents: no measurable effect within 60k steps. - A population cap: not the main cause. - Key-lock pathogens against monocultures: they do hurt monocultures, but the generalist recovers. My constraint, which I care about: no artificial hacks. I do not want to add a multiplier that "makes diversity work". If the ecosystem needs diversity to be stable, the reason should be ecological — a trade-off, a niche, a cost — not a magic number in the config. So I would love help from people who know ecology, evolutionary biology or artificial life: 1. Which natural mechanism most reliably produces stable coexistence of distinct morphologies in a simulation like this? 2. Is the "one generalist wins" outcome actually a bug, or is it what a well-mixed environment with one food type should produce — meaning the fix is spatial structure and multiple resource axes, not selection pressure? 3. Predator-prey: what is the minimum set of mechanisms (refuges, handling time, satiation, prey switching, hunting cost) that keeps cycles going instead of collapsing? There is a pinned issue with the details: https://github.com/shamichvolga/genepool/issues/1 You do not need to code to help. If you run a pool for twenty minutes and tell me what you saw — a strange body plan, a collapse, a predator that learned something — that is already data I do not have. Thanks for reading. Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.