Sep 20 2010

AQUAIRUS was my Junior year project at DigiPen and by far the coolest.

The project was designed from the ground up as a learning AI bonanza in the guise of a 3D, networked fish tank simulator. Rather than developing plain, stupid, AI for the fish via FSMs or some such we aimed for something more interesting. Our fish brains were composed of multiple objects chained together, each of which feeding impulses to yet another module. This module weighed and evaluated each input and sent final decisions to the fish’s body for interpretation. Each of the modules (the ones that generated the impulses and the ones that translated those into a decision) were implemented with one of three machine learning algorithms.

For a given tank first batch of fish were plopped into the tank with randomly assembled brains. The fish would either survive, or not. Successful fish survived long enough to produce offspring which would inherit a similiar brain structure, while unsuccessful fish would simply die out.

The end result of all of that was a tank full of fish with brains (code-wise) far more complex than something done a plain FSM. Sometimes these fish would have exibit some genuinely interesting behaviors, stuff right off the Discovery Channel, but typically we’d spin the simulation just to get a tank full of very stupid fish.

Still it was an fun expirement, even if it would have benefited from another semester of our time.


Team Roster

Screen Shots