This is the fourth and most advanced version of my agent-based evolution simulations for ALife (artificial life). I'm not really sure where versions 2 and 3 are anymore, but you can see version 1 here. All I have left of version 4 is this 20 second recording which I took after leaving the simulation to evolve for what I assume was several hours.
The purpose of this simulation was to use a genetic algorithm to evolve intelligent agents that seek food in order to survive and reproduce. The genotype of a given agent encoded a spiking neural network, producing a phenotype that changed the behaviour of how they would seek food given their 3 sensory neurons (their 3 "whiskers"). Notice the visualisation of one of the agent's brain in the red circle in the bottom left. It is similar to a typical feed forward neural network, but with an additional temporal factor due the spiking nature. The closer an agent whisker is to food, the more frequent a sensory neuron would spike.
I could write many many words about this project and the lessons I learned to make effective evolution simulations. If someone cares I'd be happy to brain dump one day.
The original inspiration for this project (and a significant part of what got me interested in AI in the first place) was this random video I found on YouTube. It was only many years later did I discover this was actually a video from a much younger Andrej Karpathy.
As framework-less web was the only stack I knew at the time, this simulation was build using raw HTML/CSS/JS using the Canvas element to render graphics.
I was super passionate about this software and I'd love to make more versions in the future and I have many plans for how I would extend it. I'm also a much better software engineer now so I think I do a lot better. Unfortunately I've never really given myself the time as it's not particularly useful. Maybe one day.