/lasertag-v0

Implementation of Deepmind's LaserTag-v0 game in A Unified Game-Theoretic Approach to Multiagent Reinforcement Learning(2017)

Primary LanguagePython

lasertag-v0

This is an implementation of Google Deepmind's LaserTag-v0 game in A Unified Game-Theoretic Approach to Multiagent Reinforcement Learning based on deepmind/pycolab.

Install

cd lasertag-v0
pip install -e .

How to use

import gym
import lasertag

env = gym.make("LaserTag-small2-v0")
(p1_state, p2_state) = env.reset()

action = {"1": 0, "2": 3}

(p1_next_state, p2_next_state), reward, done, _ = env.step(action)
...
  1. For env.step, action should be dictionary like above example.
  2. State consists of both agents' partial observation state as a tuple with size 2
  3. Reward is np.array([0, 0]) or np.array([1, 0]) or np.array([0, 1]). If agent '1' wins, np.array([1, 0]) else np.array([0, 1]).

LaserTag-small2-v0

small2

LaserTag-small3-v0

small4

LaserTag-small4-v0

small4

Contribution

I'll appreciate any help, issue, pull request. Thanks!