The BombSolver has no knowledge of the game code. It only controls the cursor and identifies features on the bomb by taking screenshots.
It starts by rotating the bomb to capture images of all six sides. It then identifies all the modules and features of the specific bomb (does it have any indicators, how many batteries are there, what is the serial number, etc.).
It then goes through each module/puzzle and uses image analysis / machine learning to identify the different parts of the specific puzzle. The solutions from the bomb manual (which are hard coded into the program) is then used to actually solve the modules (again, simply by controlling the cursor).
OCI, using a seperate Neural Network, is used to solve the Who's On First?, Memory Game, Button, and Password modules (as well as for some other miscellaneous things).
A seperate network is used for solving the Symbols module.
The Maze module is solved by first detecting which maze it is, and then finding the optimal path using BFS.
The Password module uses a DFS-ish approach, in order to quickly find the most likely word.
Any (non-modded) bomb can be defused in record time!
