I’ve got a racing game with a handful of AI racers. They’ve been trained prior, and have a perfectly functional behavior model. Their AI components stay disabled during the race’s initial countdown, then are enabled when the race begins. This triggers a request to initialize the Academy singleton, which freezes the game for a few seconds.
I may be misinterpreting the role of the academy, but do I even need it to be initialized if my agents already have their model? If it’s not needed, how do I prevent them from trying to initialize it?
Alternatively, in the case that it is needed, I would like to be able to initialize the academy when the scene is loaded, before the agents are enabled.
Hi @Chippybippers ,
The academy tries to search for an external python process for training first before going into inference mode. We have some changes coming that will allow you to configure the academy to not look for the training process, which should speed up your startup time.
The Academy is the driver of the ML-Agents event loop at the moment. It steps all of the Agent Components so that you can have a happy ML-Agents loop. This is also something we have long-term thoughts about changing to make it more customizable.
Are you seeing this in the editor, or in a Player build, or both?
It definitely occurs in the editor, and it looks like it does in a player build as well. It sounds like the changes you have coming will deal with my problem, though.