Overmind

This project is abandoned and its default file will likely not work with the most recent version of StarCraft II. Whether this project is out of date or its author has marked it as abandoned, this project is no longer maintained.

Overmind

Overmind will be a non-cheating, decision making AI for Zerg in Starcraft II.

Overview

Initial goals:

  • Intelligent scouting
  • Time and place of attack determined by decision-making based on in-game conditions
  • Unit composition determined by in-game conditions
  • Build orders calculated on the fly based on in-game conditions
  • Adjustable personality variables for setting aggression levels, etc.
  • No cheating to include:
    • No harvest bonus
    • No map vision
    • No inappropriate use of information about enemy players.

Successes

  • 3/23/2010 Overmind is capable of using the CIA features to determine the location of the enemy's Main and send an Overlord to monitor production.
  • 3/18/2010 - Overmind defeats Starcrack Protoss AI OpenGnd1 (19 Zealots) in 2 different 1v1s.
    • Overmind is still using Stock() build orders etc.
    • Game outcomes are still largely random.
  • 3/17/2010 - Overmind uses a simple unit comparison algorithm for Zerglings and Banelings vs Zealots to determine unit composition for attack waves and timing of attack. Successfully understands to send the next wave quickly when enemy force levels are low.

Data Collection

A sub-section that focuses on gathering information.

Unit Choice

A sub-section that focuses on selecting what unit composition to build.

Production Management

A sub-section that focuses on building, training, and determining build order.

TODO

Have the AI successfully generate and execute a starting build order such as 12-Pool.

Cerebral Intelligence Agency (CIA)

A sub-section for scouting/intelligence gathering.

Tactical

A sub-section for improving unit and wave AI

Chat Debugger

A sub-section for chat command based debug functions. Suggested by vjeux.

Problems

  • Overmind is still using Stock() build orders.
  • Use of (poor) default scouting algorithms leaves some expansions undiscovered and unattackable in most games.
  • Baneling tactical AI wastes them on armored units.
  • Army reponse to enemy scouts has a large and somewhat random impact on game outcome by creating unintended skirmishes.

Upcoming Milestones

  • Successfully scout out-of-the-way expansions.
  • Calculate and produce an initial build.
  • Consistently defeat Starcrack 6.1.1 Protoss in 1v1s using only Zerglings and Banelings.

Comments

Posts Quoted:
Reply
Clear All Quotes

About This Project

  • Project ID
    22603
  • Created
    Mar 24, 2010
  • Last Released File
    Never
  • Total Downloads
    0
  • License

Categories

Members

Recent Files

No files uploaded yet.