Ball Sort Puzzle with reinforcement learning

• Project name: Ball Sort Puzzle
• Short description: Solve the Ball Sort Puzzle using reinforcement learning
• Environment: Unity
• Tools: C#, reinforcement learning
• Institution: FEUP
• Course: IART (Artificial Intelligence)
• Project grade: 19.5/20.0
• Group members:
  • Diogo Miguel Ferreira Rodrigues ([email protected] / [email protected])
  • Rafael Soares Ribeiro ([email protected])
  • Bernardo António Magalhães Ferreira ([email protected])

Based on game Ball Sort Puzzle by Spica Game Studio.

Installing and running

1. Open BallSortPuzzle folder as a Unity project (v. 2020.2.2f1)
2. If scene is empty, open the Scene file, in the Scenes folder
3. Select the Board GameObject in the hierarchy, and in the Behaviour Parameters script, select a model brain to test.
4. Make sure the Behaviour Type is set to Inference Only (this is so that ml-agents uses the model to make decisions, instead of training by default)
5. Click play to see the model play some random, simple ball sort puzzles. After every game, a new one of the same proportions and difficulty is generated

(To manually play the game, select Heuristic Only in the Behaviour Type, play with the mouse, selecting the tubes to move the ball from and to with LMB. Note that due to the way ml-agents polls for input, some mouse inputs are lost)

License

© 2021 Rafael Ribeiro, Diogo Rodrigues, Bernardo Ferreira

All files are licensed under GNU General Public License v3 by © 2021 Rafael Ribeiro, Diogo Rodrigues, Bernardo Ferreira, to the exception of:

• all files inside directory doc/report and all artifacts deriving from them, which are licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International public license.