tjc-gym 0.1.0

OpenAI Gym environment of Traffic Junction with continuous action space

tjc-gym

Traffic Juntion Continuous (tjc) is a continuous version of the environment Traffic Junction defined in Learning Multiagent Communication with Backpropagation. The environment is based on OpenAI Gym and has an continouos observation and action space where the original is discrete.

Installation

Through PyPi:

pip install tjc-gym

Or directly by cloning repo:

git clone https://github.com/jakobdybdahl/tjc-gym.git
cd tjc-gym
pip install -e .

Usage

import gym

env = gym.make('tjc_gym:TrafficJunctionContinuous6-v0')
done = [False] * env.n_agents
score = 0

obs = env.reset()
while not all(done):
  env.render()
  actions = [acsp.sample() for ascp in env.action_space]
  obs_, rewards, done, info = env.step(actions)
  score += sum(rewards)

Reference

If you want to site our work, please see the citation file in the repo for details. Through GitHub it is possible to 'Cite this repository' and get a BibTeX reference - it can be found in the 'About' section of this repo.

Acknowledgement

This environment was developed by RasmusThorsen and jakobdybdahl to complement our master thesis at @ Aarhus University.