We wanted to apply a growing technology to something fun without worrying too much about the prizes. Visual asteroid recognition and avoidance may not apply to the banking, analytics, or the transport industry today, but following the first car in space who knows what the future holds.
What it does
The algorithm takes in a 2D mapping of the world. This is then converted into a 2D detection array based on material or data reads. This is the AI input. The AI is mutated and evolves through evolution, eventually making sense of our detection array. The AI's fitness is based on distance traveled and coins collected. A collision with the side barriers or an asteroid ends the journey short.
How we built it
We built everything from scratch. The most complicated data structure we pulled from somewhere else was a List. Our data prep and network implementation is all in C#. The game building / AI simulation is done in Unity.
Challenges we ran into
We didn't intend for out main focus to be on a self made AI. This made for a major crunch at the end.
Accomplishments that we're proud of
The way we convert known materials and tags in cells to our 2D detection array. Also our pipeline into our network.
What we learned
Network implementation strategies.
What's next for Flappy AI
1- Allow the AI to learn from player. 2- Different AI learning schemes (linear regression etc.)