Sample-efficient Neurosymbolic Proximal Policy Optimization

This repository hosts the code for the paper Sample-efficient Neurosymbolic Proximal Policy Optimization.

Repository layout

Folder	Description
gym-subgoal-automata/	PPO and neurosymbolic variants (H-PPO-Product, H-PPO-SymLoss, H-PPO-SymLoss-Eps, H-PPO-SymLoss-Theta) on the OfficeWorld and WaterWorld domains.
minigrid/	PPO and neurosymbolic variants (H-PPO-Product, H-PPO-SymLoss, H-PPO-SymLoss-Eps, H-PPO-SymLoss-Theta) on the MiniGrid-DoorKey domain.

Refer to the README.md file present in each subdirectory to install the respective virtual environments and run the experiments.