GymRemoteRenderer

Remote Renderer for OpenAI Gym. This code for solving a problem as below line to run open ai gym application on remote system like AWS.

pyglet.canvas.xlib.NoSuchDisplayException: Cannot connect to "None"

Preparation

Need to install packages.

pip install numpy zmq pillow

Need to install xvfb-run application.

sudo apt-get install xvfb

How to use

Server

Run pystream.py on local system. If you want to change port, run with argument -p .

python ./pystream.py -p 5555

Client

Import RemoteRenderer from pystream to your code.

from pystream import RemoteRenderer

And create instance by with syntax. If you want to changel IP address and port of server, create instance with address and port parameters.

with RemoteRenderer(env, address='192.168.0.101', port=5555) as renderer:

Example

import gym 
import numpy as np
from pystream import RemoteRenderer

env = gym.make('CartPole-v0')
with RemoteRenderer(env) as renderer:
    env.reset()
    for _ in xrange(100):
        values = env.step(env.action_space.sample())
        renderer.render()

Run script

To avoid pyglet.canvas.xlib.NoSuchDisplayException, run on virtual frame buffer by xvfb. You can run script using xvfb.sh easily.

./xvfb.sh [script file]