LunarLander-v2 using image

lunarlander_v2_image.py

import gym
import numpy as np
from gym.spaces import Box


class ValidateLunarLanderImageEnv(gym.Env):
    def __init__(self):
        self.env = gym.make('LunarLander-v2')
        obs = self.reset()
        self.observation_space = Box(
            low=0,
            high=255,
            dtype=np.uint8,
            shape=obs.shape)
        self.action_space = self.env.action_space

    def convert_obs(self, observation):
        observation = np.atleast_3d(observation)
        if observation.max().item() != 0.:
            observation *= 255.0/observation.max()
        return observation.astype(np.uint8)

    def reset(self):
        observation = self.env.reset()
        return self.convert_obs(observation)

    def step(self, action: int):
        obs, reward, done, info = self.env.step(action)
        obs = self.convert_obs(obs)
        return obs, reward, done, info

    def close(self):
        self.env.close()
        return self.close()

Hi @ipsec ... my comment is unrelated to this file. I'm trying to get your solution to this stochastic muzero nan training problem which was closed on another repo. Can you please share your solution? Thank you very much
DHDev0/Stochastic-muzero#2 (comment)