Gym Routing: Reinforcement Learning Environments for Routing Problems

Gym-Routing aims at offering a suite of Gymnasium environments tailored for training Reinforcement Learning (RL) agents to tackle various routing problems.

Table of contents

• Installation
• Key features and improvements
• Tutorials and RL Agents Implementations
• Repo setup

Installation

1. Pip install the package gym_routing defined in the repo

pip install git+https://github.com/PierreCounathe/gym-routing

2. Import gym_routing and make the gym environment

import gym_routing

import gymnasium as gym
from gymnasium.wrappers import FlattenObservation

env = gym.make("gym_routing/TSP-v0")  # use "gym_routing/ActionMaskedTSP-v0" to use the action masked environment
env = FlattenObservation(env)

Key features and improvements

Environments

• (TSP) Vanilla Traveling Salesman Problem
  • Action Masked TSP
• (VRP) Vehicle Routing Problem
• (CVRP) Capacitated Vehicle Routing Problem

Tutorials and RL Agents Implementations

Basic usage with Stable-Baselines3 algorithms implementations

import gym_routing

import gymnasium as gym
from gymnasium.wrappers import FlattenObservation
from stable_baselines3 import PPO


# Define the environment
env = gym.make("gym_routing/TSP-v0")
env = FlattenObservation(env)

# Define and train the agent
ppo = PPO("MlpPolicy", env, verbose=1, tensorboard_log="./ppo_tsp_tensorboard/")
ppo.learn(total_timesteps=5_000, progress_bar=True)

Repo setup

To ensure clean and consistent code, this repo uses pre-commit hooks (black, flake8, mypy, isort), that can be installed the following way:

pip install pre-commit
pre-commit install