A Game AI Project Using Unity ML-Agent
Unity3DML-AgentDeep Reinforcement LearningProximal Policy OptimizationAIMaze GameGame AI
Important Notice: The Game AI only work on MacOS for now.
We would like to design a game that suits comprising AI player. A maze game is created in this project since there are only two results, namely, win and lose. Only two final results can be easily simplified to rewards in the reinforcement learning algorithm.
A maze game is designed based on a prototype JumpWall game got from GitHub Repo of Unity ML-Agent.The final goal of player is that moving the cube, which represents the player character, to the escape point (destination). The player will have a first-person perspective, this will increase the difficulty of escaping from the maze.
Besides, in order to increase the player’s feeling of tension, we design a competition mode. The player will compete with a AI player. Once there is a party reaching the destination, the whole game ends. The one who reaches the destination win the whole game.
To give player more entertainment and make one turn of game has reasonable time period, we designed 15 rounds in total per turn. The player has to continuously finish 15 rounds (reach the destination 15 times) before they win the game.
The ML-Agent Use PPO for Algrothim.
You can also check our report in the Report&PowerPoint directory for more detail.
- MacOS: 10.14.1
- Unity: 2018.2.18f1
- Visual Studio: 7.6.11
- Python: 3.6
- Tensorflow: 1.7.1
- Unity ML-Agent: 0.6
- pip: 18.1
Download the Program
Run the Application MazeGame_v1.0.xx.app in the build directory.
Select the Screen resolution and click play button
WASD for Movement, Mouse for Rotate view. Restart button for the restart.
Notice: If you find that the AI does not move in your game, you have to follow the Installation guide and build the Game in your own.
$ pip install tensorflow==1.7.1git clone https://github.com/Unity-Technologies/ml-agents.gitchange to the ml-agents directory:
$ cd ml-agents/ml-agentsInstall the package in the path.
$ pip3 install .Also Follow the Official Document ML-Agent Installation Guide and ML-Agent Basic Guide to install ML Agent
$ git clone [email protected]:tavik000/MazeGameAI.gitCopy the WallJump Directory to ml-agents/UnitySDK/Assets/ML-Agents/Examples/ and Replace the original one.
Tips: The WallJump is the Scene of our Maze Game, We did not change the folder name.
Add the project to Unity from ml-agent/UnitySDK
Open it from Unity
Direct to the Scene of WallJump from Assets/ML-Agents/Examples/WallJump
You can use different Pre-trained model from our TFModel directory
For using model, you should to use the internal brain type.
Important Notice: When you want to use different model. You have to change the brain type to external first, then change back to internal. Otherwise, it will still load the previous model
Click the play button for testing. When the game start, you have to check if the Graph Model is the one you want to use or there is still the previous model.
The game is speeded up in Unity. You can build the game to local and run.
Please refer to Unity Document for more information about training.
$ mlagents-learn config/trainer_config.yaml --run-id=yourID --trainThe Maze
Training
The Reward Function
Gameplay
Gameplay Video
If you like this, please leave a star.
Email: Key [email protected]
MIT License
Copyright (c) 2018 key
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
All Copyright Reserved