If the bot can't get further after trying random jumps for 10 attempts, it will delete the last jump in the list. The only inputs it uses are the current X position, the mode of the player, and whether the level was finished. Note that the screen is not used as an input at all - the bot plays blind.
If these new clicks help it go further through the level then it saves the new clicks, and those are now used for the jump list. (More about this later.) After it does all of the jumps, it starts to randomly click. This means that overall training runs much faster than 2.0, but it may take a while if the bot gets stuck somewhere, especially near the end.īasically, the bot first follows a list of jumps that have been saved and are confirmed to work. DashBot 2.0 used a full genetic algorithm (with generations and species), but DashBot 3.0 removes the species concept and defines a new generation as one that is better than the last. Geometry Dash bot to play & finish levels - Now training much faster! How it worksĭashBot 3.0 uses a very simplified version of a genetic algorithm, which uses random evolution to slowly create better generations for future species to evolve over.
