This script simulates and records every 10th episode of the CartPole environment in OpenAI Gym, saving videos in the ./vid directory. Each episode runs until completion with random actions, then resets for the next.​⬤

record_episode.py

import gym
 
# Initialize the CartPole environment with rendering mode set to 'rgb_array'
env = gym.make('CartPole-v0', render_mode="rgb_array")
 
# Wrap the environment with the RecordVideo wrapper to record videos
# The episode_trigger lambda function ensures that a video is recorded every 10 episodes
env = gym.wrappers.RecordVideo(env, "./vid", episode_trigger=lambda episode_id: episode_id % 10 == 0)
 
# Reset the environment to get the initial observation and additional info
observation, info = env.reset()
 
# Simulate 100 episodes
for i in range(100):
    while True:
        # Sample a random action from the environment's action space
        action = env.action_space.sample()  # Agent's policy can replace this for specific decision-making logic
 
        # Step the environment using the sampled action
        observation, reward, terminated, truncated, info = env.step(action)
 
        # Break the loop if the episode is complete
        if terminated or truncated:
            break
 
    # Reset the environment for the next episode once the current episode is done
    observation, info = env.reset()
 
# Properly close the environment to free resources
env.close()