Installation
pip3 install -e .
Usage
import gym
import gym_toy
Brief description of the environments
Tabular
random_switch.py: 2D gridworld where the agent has to press a switch (5th action) in certain cells to get a reward.
Continuous state / discrete action
gridworld.py: 2D env with sparse rewards.dig.py: 2D env where the agent has to dig land (5th action) to find rewards.
Continuous control
gridworld_continuous.py: 2D env sparse rewards.lqr.py: linear-quadratic regulator.lqr_sparse.py: state penalty is always -1, except when the agent is close to the goal (distance < 1).sparse_car.py: car moving on a 1D plane with sparse reward.sparse_navi.py: agent navigating on a 2D env, with linear dynamics and sparse reward.pendulum_sparse.py: like gym Pendulum-v0, but reward is sparse.
Multi-objective
mo_lqr.py: multi-objective linear-quadratic regulator.mo_grid.py: multi-objective gridworld (the farther the reward, the higher its value).