'ai' allows to send command through env.step, 'human' allows to play with a controller on the respective port, 'cpu' is game cpu. have been tested, the others stages should work to (you can look in ssbm_gym/gen_code.py for a list). stage: string, what stage to be played on.windows: boolean, required if you use the program on windows.render: boolean, if True, show game window and set the fps as 60.exe: string, specifies wich dolphin executable to use (should be good by default if you followed the installation guide).Here is a documentation of the different options. The structure of observations monitored in the game and the definition of actions are visible in ssbm.py (it's a bit messy at the moment).
#Howntombuild 20xx melee iso how to#
This shows how to define an action space, compute a reward, and so on so you can implement your own environment. You can look in ssbm_env.py to see the abstract BaseEnv class and a simple implementation of it with the SSBMEnv class. There is a basic usage example in test_env.py. The default name should be SSBM.iso Usage Melee NTSC v1.02 that you put or link in the ISOs folder. You also need an ISO of Super Smash Bros. Windows is supported only for playing or visualizing a bot, it is not recommended for training. The compilation has only be tested on Ubuntu 18.04 at the moment.įor Windows, you should download this compiled version of Dolphin, and put the content of the 圆4 folder inside dolphin-exe. You need Python 3 for this to work (use pip or pip3 depending on your setup).įor Linux also you need to build a specific Dolphin executable from the sources located here (see instructions below). It's based the great work of xpilot/vladfi1. OpenAI gym type environment for Super Smash Bros.
0 kommentar(er)
