- Python 3.6
We recommend to run in a new virtual environment, e.g. conda create -n social-grid python=3.6. You can clone this repo and run pip install -r requirements.txt to get the dependencies.
The environment is adapt from "Help or Hinder: Bayesian Models of Social Goal Inference". The reimplementation and a GUI is in the world folder. You can check social_world.py to see how to initialize the environment and show the GUI. This implementation follows OPAI gym's interface to step and reset the environment.
- Sample a new environment config:
SocialWorldEnv.sample_init_grid() - Step with two agents' actions:
env.step(a_strong, a_weak). This returns two 4-tuples(obs, reward, done, info), the first one is for the strong agent, the second one is for the weak agent. - Resest the environment:
env.reset()