In this project AI-Actors were trained to survive in a procedurally generated Ecosystem. Using a cutom-written Neural Network with Genetic Algortihm (Scoring Function).
- Written in C++
- Neural Network in SIMD (AVX2), 8-Bit Fixed Point
- Genetic Algorithm: Crossbreed & Mutate
- Mutation Parameter via Evolution (chance, rate)
- Multihreaded Training
The simulation includes procedurally generated levels (non-exclusiv: underwater, several food types, collidable) with the actors being able to move in any direction, eat, wait or drag items whilst facing the threats of starvation and drowning.
Including a HTML/JS Web-Interface with Webserver via Crow for visualization:
create_project.bat
MSBuild /p:Configuration=Release /nologo /v:mSee Progress Report here.