alphaBot

Farming bot for the game Alpha Wars (http://eng.alphawars.com/)

Introduction

(Not important for running alphaBot)

In my last exam period, I was searching for a simple browser game that can be played without to much intellectual effort. Something for the evenings after exhausting days of learning. One of my attempts was Alpha Wars. A game in which you must produce raw materials to update your buildings and create an army. It seems that the whole purpose of this game is to wage war against other players and there is no room left for players that prefer to build and collect. Thus it became boring to me very fast. Additionally it turned out that the game follows the Pay2Win principle what makes it very hard for me to like that game at all. Something that aroused my curiosity was the captcha system of the game. To prevent the usage of bots ( ;) ), you need to click on a captcha every time you produce a raw material. The captcha consists of a number of simple geometric objects, mostly triangles and a single rectangle. To solve the captcha, you need to identify and click on that single rectangle. Then you can produce a raw material for exactly one minute. If you want to produce longer than a minute, you need Titanium. This can be bought or created by transforming fuel. But the transformation is very expensive and the costs increases with every executed conversion. To become a serious threat to your opponents you need to buy Titanium for money, otherwise you spend whole days by clicking every minute on a capture.

Example of an Alpha Wars captcha.

Clicking on a rectangle. Seems to be a very simple task, even for a computer. My plan was to take a screenshot and do some image processing. I always searched for a project where I could use image processing, because I was eager to learn the skill. Thus it was a great opportunity. My first approach was very optimistic. I thought one hour for a python script and my computer solves the captchas for me. Yeah... one hour became five hours and five hours became a couple of days. Mostly spent with optimizing the algorithms and the assessment parameters.

My first approach was to apply a corner detector and cluster the detected corner points. A problem of this approach is that the number of clusters is not fixed. Sometimes there are up to four different geometric objects, but on other occasions there is only one single rectangle. So I needed to cluster for k = {1,...,4}. This provided a big pile of clusters and I needed a rating function that founds the best fitting (and hopefully the correct) cluster. My plan was to determine the convex hull of a cluster and then compare the points within the hull with the points that are part of the convex hull itself. This alone was very inacurate and I decided to consider also the angles between adjacent lines of the convex hull. My plan was to analyse every possible clustering and define its rating according to the number of right corners (+/- an epsilon) in the convex hull. Actually, the algorithm worked... kind of. The hit rate was 4/7 that means: for 7 clicks the algorithm hit the rectangle in 4 cases. It seems not very good but if you do not have to stand beside your computer it is acceptable. However, it turned out, that the game has an additional bot protection. If you click on the wrong object to often, you will be logged off eventually. The number of wrong clicks until you are logged off is adaptive. There is a point where you will be logged off every time you hit the wrong object.

Thus I needed to increase the accuracy of the algorithm. But this was impossible by using the corner detector, because I found no way to increase the quality of feature points on a representable set of different captchas. It worked only for a few of them, while the results on the remaining worsened. The solution was already part of my thoughts, but until now I was to lazy to implement this idea: I had to analyse the lines in the image. I read some papers and found a simple and fast line finding algorithm. It is implemented in a separate library (https://github.com/highkite/lineFinding), because I am going to use it in several projects. The line finding algorithm was the answer to everything. Now the algorithm achieves a nearly perfect hit rate. And at last I am able to finish the chapter Alpha Wars.

Consider the following as my share in fighting the Pay2Win philosophy ;).

Installation

To install the alphaBot software you need to install the following packages first:

1. matplotlib
2. autopy
3. skimage
4. numpy
5. scipy

You can install them separately or by using pip and the requirement.txt file:

pip install -r requirements.txt

alphaBot is based on the lineFinding algorithm from here: https://github.com/highkite/lineFinding. Clone the repository and make sure that it is in the same directory as the alphaBot directory. That means you should have a folder with at least two directories in it: lineFinding and alphaBot.

A reasonable remark

I used an ubuntu linux system. I never tested it on Windows. I am not sure if it works correctly on windows. Especially the mouse positioning and the screenshot library might cause problems.

Usage

The usage of alphaBot is very simple. Almost everything you need to do is displayed in the console. However, in the following I provide a stepwise instruction:

1. Open a console and type python alphaBot.py.
2. Log in to the browser game and make sure that your whole base is visible in the window.
3. Press Enter in the command line and move the mouse pointer in the browser window. Wait until the Spinner disappears.
4. alphaBot is initialized now.
5. Press enter and move the mouse pointer in the browser window again.
6. alphaBot will farm fuel for you.
7. You can stop alphaBot by pressing ctrl + c.