Awesome

Computer Tennis

A gym reinforcement learning environment using the gym3 API. This is mostly a clone of the Atari Pong (Video Olympics) game, though the actual movement code isn't that close to the original. Both single player and 2 player modes are supported.

This environment has pixel observations which can be rendered with either OpenGL (which should work almost anywhere) or Cairo (if you have the python pycairo package installed). In both cases, the rendering is done off screen so there is no popup window. The Cairo version is probably faster, but harder to get working due to pycairo not having binary wheels.

The pixel observations are not square, like in the Atari game, so you have to scale to a 4:3 ratio to get the correct appearance.

In addition, X-server-less OpenGL rendering is available on Linux. To choose the EGL device, pass egl_device_index to the environment constructor.

Installation

git clone https://github.com/christopher-hesse/computer-tennis.git
pip install -e computer-tennis

Quick Start

Use the keyboard to play against the AI opponent:

python -m computer_tennis.interactive

Create the gym3 environment in code:

from computer_tennis import TennisEnv
env = TennisEnv(num=2)

To select different rendering backends, you can pass surface_type="opengl" or surface_type="cairo" to the constructor.

Create a gym environment from the gym3 one:

from gym3 import ToGymEnv
from computer_tennis import TennisEnv
env = TennisEnv(num=1)
gym_env = ToGymEnv(env)