mgmag-proj



mgmag-proj

1 1


mgmag-proj

Project in Games and Agents

On Github DrSLDR / mgmag-proj

Gravwell

  • Boardgame, 2-4 players
  • Move in direction of the closest ship
  • Play Cards to move
    • Type
    • Value
    • Name
  • Important game phases:
    • Draft
    • Play
    • Resolve
      • Emergency stop (ES)

AI-Aspect

  • Decision making
  • Learning
  • Implemented through:

    • Reinforcement Learning (RL)
    • Neural Evolution (NE)
    • Decision Tree
    • Random AI
    • No-ES Random AI

AI Statistics

Our goal: win from random AI

NB: NE does not significantly improve over random AI

AI Statistics

Interestingly:

NE holds its own against RL