It’s been a while since I’ve built anything interesting in the Artificial Intelligence field and with all of the neural network / transformer / LLM nonsense going on right now, I wanted to go back to one of the parts of AI I love, swarm systems.
If you haven’t heard the term before, boids are a generic model for the way organisms swarm together. Think fish in a school or birds in a flock. Each animal has a pretty simple set of things its trying to do but sometimes the overall effect can be breath taking; think of a huge murmuration of starlings at twilight. Boids -- think they fly like a boid -- are a way to model how bunch of simple agents can create a very complex behaviour, and how small tweaks to an agents behaviour can change the way a whole system performs (think starlings compared to geese). They're a very old addition to the world of artificial intelligence, and they've also been extreamly important to the world of film making.
 |
| Walter Baxter / A murmuration of starlings at Gretna / |
That being said, it’s been a very long time since since I’ve messed around with swarms (or boids) and I’m quite rusty and -- as I think is a theme here -- my vector math always seems to be just a touch insufficient. So, I’m going to write a short series as I go about putting my swarm system together.
I’m not going to go into all of the details right now, but I’m going to roughly follow Craig Reynolds Boid model. A lot of the details aren’t in the original paper and the content in the Wikipedia article is slightly out of line with my memory, so I’m going to bushwack from what I have to a hopefully working swarm system (rather than do more reading right now).
For this doodle, I’m taking my pre-existing moving agent code -- which I always wrote with an eye to doing a swarm system and expanding on it. To start I have little circles that try to balance being between the mouse and the middle of the screen. To build boids we’re going to have to expand that a little and add in a few more things to do.
I’m trying to stay close to Reynolds model for now, although there’s lots of fun to play with later on. I’m going to set up three “urges” for the boids, a separation urge -- “I’d like to not crash”, a heading (or alignment) urge -- “I want to go where everyone else is going” and a cohesion urge -- “I don’t want to be the only one out where I can get eaten”.
I’ve been working on this for a while -- and then not working on it for a while, so I’m starting to write with the intention of kicking myself into finishing the project. I’m about halfway done putting everything together -- which you'll notice when you look at the code in the repository, and I think I’ll write this up in four or five posts over the next few weeks. Once I get the basics done, I’m looking forward to all the other things I can play with.
