Ok so, I’ve done a loooot of different prototypes for RTS games, each time having to restart from scratch due to significant advancement in my own personal skills between each (and also a healthy case of being away from a project for too long). Most of them are very old, though, and the only stuff I have to show for it are a few gifs.
Here’s a few of an “ant ai” prototype I did way back in like 2019 or something.
The basic gist of this one was trying to use a combination of blueprint-based AI brains w/ “ant pheremone” signals and blackboarding to get an autonomous ant colony to sort of auto-mine.
The little yellow vfx are “path” pheremones, which the ants will follow if they have no targets, and the pink ones are “food” pheremones, which the ants will remember the position of, and also prioritize over following paths.
If you watch closely at the top middle part of this gif, you’ll see two ants pass each other, quickly turn around to smooch one another on the butt, and then keep going. That’s how they exchange information between one another; this way, ants can get info about where food is from scouts without having been there.
To add yet more complexity, I made a “refinery” objects (the rectangle) that would turn purple (inedible) cubes into orange (edible) ones. If an ant found a purple cube source, but didn’t know the location of a refinery, they’d leave the cube alone, but store the knowledge of where it was. Eventually if another ant found a refinery, the knowledge of where to take those cubes would slowly disseminate throughout the colony.
This was my first prototype, and was really just an experiment to familiarize myself with AI state machines and the blueprint-based way you define them.
My next notable prototype went much more in depth to the actual mechanics of an RTS, including the player controller, unit selection and movement, hotkeys, grouping, and so on. The big thing I wanted to figure out here was how RTS player controllers worked, and how the unit AI should interact with commands given by the player.
Unfortunately, I don’t really have much saved footage of this prototype, save for a few process videos. Here’s a link to some footage of an early draft of this project: https://drive.google.com/file/d/1s_Wpeym2OcDAkTzohaxS0U8EhSmz43C_/view?usp=sharing
Here’s a gif of a later version.
Link to the full video: https://drive.google.com/file/d/1hKr-xGuuhkSrV3vsW2FXd1AxDk8Hye_y/view?usp=sharing
In this, I’ve got drag-select working, some grouping / building hotkeys, and a bunch of different workers autonomously mining. The cool thing about this one I got into the weeds on was the differences between “miners” and “transporters”; in the example, the green rectangles can pick up mined resources, and move much faster than normal capsule workers, but cannot mine the raw resource themselves. You can see that all they do is try to find a full stockpile to transport. Also, all the AI brain flow stuff here is hand-coded, rather than relying on stock blueprints.
The way I implemented this was with the concept of “certifications”, not unlike real construction tool certs. The capsule miners have certs for mining resources and transporting them, but the rectangles only have certs for transporting (albeit with higher movement values), so they can’t interact with the raw resources at all. I had bigger plans for implementing combat in a manner like this, where other enemies that are attackable require a “combat” cert, and the nature of your cert determined how you attacked. All of this would allow me to re-use most of the AI brain flow for finding / prioritizing targets, and make all AI unit pathfinding / target tracking work the same way. I think it’s a neat concept.