- Readme
- Project Description
- Test Plan and Results
- User Manual
- Final PPT Presentation
- Final Expo Poster
- Self-Assessment Essays
- Summary of Hours and Justification
- Budget
There have been many recent advances in game-playing AIs, such as the Dota2 AI and AlphaGo. With this project, we explored the use of conventional and cutting edge ML techniques to create a self-learning Starcraft II (SC2) AI agent capable of defeating Blizzard's Very Easy AI.
We are scoping our project in the following manner:
- the agent will play as Terran
- against a Very Easy Terran AI provided in the retail version of the game
- on the maps Simple64 and flat64.
Each member wrote their own assesment essays.
| Work | Time | Team Members |
|---|---|---|
| Fall Semester | ||
| Learning the basics of Reinforcement Learning | 25 Hours | Kyle Arens, Jon Deibel, Ryan Benner |
| Understanding C++/Python API | 10 Hours | Jon Deibel, Ryan Benner, Kyle Arens |
| Creation of baseline AI/First RL based AI | 10 Hours | Jon Deibel, Ryan Benner |
| Documentation | 10 Hours | Ryan Benner, Kyle Arens, Jon Deibel |
| Reading + Understanding AlphaGo | 5 Hours | Ryan Benner, Kyle Arens, Jon Deibel |
| Spring Semester | ||
| Model Improvements to AI | 50 Hours | Jon Deibel |
| Running Test Cases/Debugging | 25 Hours | Kyle Arens, Ryan Benner, Jon Deibel |
| Senior Design Expo | 4 Hours | Kyle Arens, Ryan Benner, Jon Deibel |
| Senior Design Poster | 25 Hours | Ryan Benner, Jon Deibel |
| Final Design Report | 4 Hours | Ryan Benner, Kyle Arens |
| Self Assessments | 2 Hours | Kyle Arens, Ryan Benner, Jon Deibel |
| Test Plans | 6 Hours | Kyle Arens, Ryan Benner, Jon Deibel |
| User Docs & FAQ | 10 Hours | Kyle Arens, Ryan Benner, Jon Deibel |
There have been no expenses to date.