This project is an artificial intelligence designed to play the game "snake". The game is also playable so you can try it out for yourself. The AI can be activated at any point during the game and turned off in the middle of the game as well. It is designed to be able to reason about the best thing to do in any situation. Sometimes the best it can do is die.
DEMO: http://sunjay.github.io/snake/
Progress & Bugs: https://github.com/sunjay/snake/issues
On a more technical level, this AI is capable of dynamically finding the path that the snake should take to get to the goal (food) while its body is constantly moving and changing position. This is all done in real-time. Decisions must be made within about 50 ms or else the snake will likely die.
The system is designed to simulate a player playing the game as closely as possible. To accomplish this, the system is designed with two processes.
The main process continously runs and moves the snake with no regard for whether the snake is being controlled or not. The main process renders the board on the screen and runs at about 15 FPS.
The ai process must send messages regarding the directions the snake on the main process should turn in order to survive and collect food. The main process keeps the AI up to date on when it updates and where the goal currently is. If the AI does not send its messages in time, the snake on the main process will likely die. The AI process does not have access to the new goal until the main process generates it.
15 FPS is ~67 ms/frame. There is an additional latency to be accounted for between the processes. Removing that and also an additional margin of time to account for any other latency or further message backlog results in a maximum of 50 ms to 60 ms of time left for processing.
In order to meet these time requirements, the entire algorithm is not executed at once. Instead, it is completed in stages in between which there is time for messages to be sent to the main process.
This approach allows for about 3 * 60 ms = 180 ms of time for the entire algorithm to take place.
The first stage completes in one frame which is the time it takes the snake to move between tiles. At that point, the snake is safe to continue on that path until it reaches that goal. Once this stage is complete, the AI process begins sending directions to the snake.
The second stage is designed to provide enough time for the third stage. In this stage, a single directional step is added to the path in order to give the snake one more frame after it reaches the goal. The goal is just to "not trap" the snake. That means giving it a way to get out of its situation if it needs to. So long as the snake has freedom, it is usually possible for it to get to its next goal. This second stage is only activated once the snake takes the first step of the path planned in the first stage. This stage must take less than a single frame of time.
The third stage can only begin once the snake has reached the goal. At that point, a new goal becomes available. The snake should take the additional frame alotted by the second stage to figure out a path to the next goal that is now available.
Run npm install to install the dependencies.
Run npm start to run the development server. This will automatically hot reload your code when it changes (with some limitations).
Go to http://localhost:8080 to see the app running.
You can run npm start and go to that address anytime now to see your code. (You don't need to run npm install every time)
Building:
Run npm run build to compile all necessary files in a dist folder.
When just running the development server. The bundle and other dist/ files are kept in memory. Use the build command to explicitly put them there. This will be done for you when you deploy so you don't need to ever really do that unless you want to see the built result.
Deploying:
Run npm run deploy to checkout gh-pages (created if not already there), merge master, build the code, commit and push the generated bundle.
Deploy will use git subtree push as described in the article Deploying a subfolder to GitHub pages.