snakai 有AI的贪吃蛇

snake is running! our project final goal is to build a snake game driven by Reinforcement Learning.

贪吃蛇出动！远期目标是实现强化学习驱动的贪吃蛇游戏。

Setting / 设置

language: Python3

package-name: snakai (snake + ai)

Start / 开始

# create env and active it
python -m venv snakai_env
source snakai_env/bin/activate
# install 
pip install snakai
# or `pip install -e .` in dev mode
# run snake using QLearning based strategy.
run_snake -s qlearning 
# `run_snake -s rule_based` to run in rule-based strategy, or `-s manual` to playing the game...
# `run_snake -h` to see all params

Attention: if you want to run qlearning based strategy, you need train it firstly.
Just go to qlearning-strategy and run the scripts' script to train the model.
Currently, about 13 minutes is needed.

you will see following screen if no exception.

screen by the fantastic ScreenToGif

Progress / 进度

• draw snake based on curses. 绘制贪吃蛇（控制台） ✅✅✅

  从 https://gist.github.com/sanchitgangwar/2158089 下载了一个示例，照着学习。

  • 完成 package 构建 ＆ 配置上传到 pypi

• rule-based strategy. 基于规则的AI ✅🔲🔲

  • 完成一个基于规则的策略。目前看比较笨。之前调研的时候发现有用 哈密顿连通图 之类来做，后续可以试下。
  • 另外，还准备试下 A* 算法，之前说好要在博客介绍一下 A* 算法的

• RL-based strategy. 基于强化学习的AI [little milestone] ✅🔲🔲

  • 目前做了基于 Q-Learning 的策略。具体见 qlearning-strategy.
  • 后续准备优化 & 尝试 Deep Q-Learning

• unity or some other game-engine driven snake? 基于其他游戏引擎的贪吃蛇？ 🔲🔲🔲

日志

roadmap

• 2021.08.08: 完成 qlearning 的基础版本。到此，最基础的策略都已经实现了！
• 2021.01.03: 再次重新组织代码结构，完成了策略逻辑和绘制逻辑的拆分
• 2019.12.28: 有空重新组织了代码结构，完成了绘制逻辑和状态逻辑的拆分；写了snake游戏的开发关键