Star-MaNN

This is an experimental combination of an arcade-style game with human supervised AI learning made for the AI and Games Jam 2021 in Pico8. The AI start with a simple set of rules, and may not seem too smart. But over time, you can teach them new rules which bring out new behaviors.

The enemy AI is a very simple Neural Network inspired by the ghosts in Pac-Man. The AI cannot learn on its own, however, by setting the AI with your favorite behavior as the new parent, you can teach the enemies to be smarter (or not!). Random changes to the parent AI (through new genes and mutations) lead to new enemies that behave differently. If the new enemies seem smarter, you can set them as the parent to continue training your own enemy AI. You even have control over how much each new AI changes from the parent with two probability settings in the menu.

After training your AI, you can see how many pellets you can collect against them. But as you collect more pellets, the enemies get faster!

Controls:

Left/Up/Right to move your ship

X while playing to automatically generate a new AI from the parent

C at any time to set the current AI seed (shown at the top) as the parent

Options:

New Gene Probability: likelihood of new genes being passed on. These new genes generally create viable AI even when this parameter is high (near 1).

Mutation Probability: likelihood of random mutations occurring. Lots of mutations can cause buggy AI, but some mutations are needed to get new and interesting results.

Show enemy targets: shows the location that each enemy is moving towards. This can be interesting to learn how the AI is working, but takes away some of the challenge in going for a high score.

Sound on/off.

Background graphics (stars and meteors) can be turned off if they are distracting.

AI seeds can be set directly in the menu (keyboard inputs should work too: e.g. type "F" and the cursor location should become F). Here are some seeds you could look at. Feel free to share some of your favorites as well:

00B005-2C0004-0000D5-1B1170

000000-0000E4-B10100-0000ED

0000F0-0000F0-0000F0-0000F0

0B0005-B00005-B11170-000000