programming

Evolving Virtual Creatures 2018-11-30

Research

• Using simplicity bias to guide evolution
• Genetic Representation

Attributes

• Cold resistance
• Heat resistance
• Simulation priority
• Reproduction penalty time ( = sum of all settable attributes )

Inputs

• Potential mate pearson correlation coefficient (primary matrix)
• Temperature (between -1 and 1)
• Light level (-1 to 1)
• Crowdedness (0 to 1)

Outputs

• Reproduce
• Move north
• Move east
• Move south
• Move west

DNA

• Two matrices
  • one from Inputs + Attributes to vector with length N
  • one from length N to Outputs
• Crossover is done over Attributes and two matrices
• Mutation is done with low probability on primary matrix, medium probability on attributes and high probability on secondary matrix
• Only creatures with same length N can reproduce with each other
• Potential mate is computed by selecting closest other creature with length N

Simulation

• Rectangular grid, split in subrectangles. Loops horizontally
• Each step, y subrectangles are chosen, with higher probability at the north and south pole
• For each chosen subrectangle, x times, choose a creature and simulate a step
• Temperature simulation is done with a linear falloff each step, temperature is added according to sunlight on a sphere
• Each step sun angle is updated