BeamNG.gym

BeamNG.gym is a collection of Gymnasium environments that cover various driving tasks simulated in BeamNG.tech.

Installation

Standard pip can be used to obtain the package of environments:

pip install beamng.gym

Or install the version from source by:

git clone https://github.com/BeamNG/BeamNG.gym.git
cd BeamNG.gym
pip install --editable .

A copy of BeamNG.tech is also required to actually run the scenario. The basic version is freely available for academic non-commercial use.

This version is compatible with BeamNG.tech 0.30 and BeamNGpy 1.27.

Configuration

The environments assume an envirionment variable to be set that specifies where BeamNG.tech has been installed to. After obtaining a copy, set an environment variable called BNG_HOME that contains the path to your local installation's main directory -- the same that contains the EULA.pdf file.

Usage

BeamNG.gym registers the environments with the OpenAI Gym registry, so after the initial setup, the environments can be created using the factory method and the respective environment's ID. For example:

from random import uniform

import gymnasium as gym
import beamnggym

env = gym.make('BNG-WCA-Race-Geometry-v0')
env.reset()
total_reward, done = 0, False
# Drive around randomly until finishing
while not done:
    obs, reward, done, aux = env.step((uniform(-1, 1), uniform(-1, 1)))
    total_reward += reward
print('Achieved reward:', total_reward)
...

Environments

Currently, the only environment is a time attack on the race track in the West Coast USA level of BeamNG.drive. New environments are being developed.

WCA Race

In this setting, the car spawns at the starting line of the race track in West Coast USA and has to race one lap. A detailled description of the observation and actions can be found in the documentation of the respective class WCARaceGeometry.