An OpenAI Gym environment to conduct reinforcement learning experiments in RCT2 track construction.
Run trackrl.py to train a model. Then run tracknn.py after specifying the model checkpoint path on line 28.
Python library requirements:
- pyobjc-framework-Quartz (for screenshots - Mac only)
- pyzmq
- gym (OpenAI)
- numpy
- dill
- tabulate
- dm_tree
- ray[rllib]
- tensorflow
- tqdm
- matplotlib
OpenRCT2:
- Must use my modified version at https://github.com/campbelljc/OpenRCT2.
- Building requires installing the ZMQ library (https://zeromq.org). There may be some issues getting it recognized by the OpenRCT2 build process...
- To run, cd to the OpenRCT2 build directory, then run the command to launch the game (e.g.,
./OpenRCT2.app/Contents/MacOS/OpenRCT2), and start a blank scenario (e.g., one of the two in the scenarios folder from this repo).