Evolutionary Agent Simulation

A simple genetic-algorithm & neural-network based agent simulation on a 2D grid.

Key Concepts

• Genome

  • Represented as a fixed-length hex string (GENOME_LEN = WEIGHTS_COUNT × 4 hex digits).
  • Each 4-hex group → one 16-bit unsigned integer → weight in [–1, +1].

• Neural “Brain”

  • Inputs (SENSORS):
    • 8 cells in ring 1 + 16 cells in ring 2 around the agent
    • Age (normalized), X/Y position (normalized), random bias
  • Architecture: 4 fully-connected layers of size 16 → tanh activation → 4 outputs (move N/S/E/W).

• Mutation & Reproduction

  • Mutation rate = 2 / GENOME_LEN per hex digit per generation.
  • Survivors chosen by spatial criteria; next generation sampled (with mutation) from survivors.

• Selection Modes

  • Center: agents inside a circle at grid center
  • Corners: agents inside quarter-circles at each corner

• Outputs

  • Runs for GENS generations; logs survivor counts.
  • Saves per-generation GIFs & final survivor-count plot in output/.