MDP State Doesn't Change After First Action

18 views

Skip to first unread message

Andrew Kvapil

unread,

Oct 18, 2017, 5:53:27 AM10/18/17

to BURLAP Discussion

Hello, BURLAP developers.

Please pardon my ignorance in case this question has already been answered elsewhere. I'm writing a BURLAP MDP and all seems to be going fairly well, except that for some reason, the state obtained from FullStateModel.sample() doesn't seem to be "applied", so to speak. The very first action seems to be processed correctly. Future ones fail to mutate the current state, however. I'm using IntelliJ IDEA. The debugger revealed that FullStateModel.sample() is being called correctly, but always with the same (outdated, initial) state.

I'm sorry not to include any code in my first post, but BURLAP is fairly modular and, being new to the library, I have a hard time analysing exactly which classes are relevant to this issue. I'm asking whether anyone has had a similar problem or if it couldn't be a BURLAP bug. I'm on version 3.0.1.