Skip to content

Latest commit

 

History

History
34 lines (21 loc) · 1.3 KB

README.md

File metadata and controls

34 lines (21 loc) · 1.3 KB
Raw

A3C for Kung Fu in OpenAI Gym

This repository contains code for training an A3C agent to play Kung Fu MasterDeterministic-v0 environment in OpenAI Gym.

Key Features:

  • Implementation of the A3C (Asynchronous Advantage Actor-Critic) algorithm for multi-agent training.

  • Preprocessing pipeline for Kung Fu observations using the PreprocessAtari wrapper.

  • Environment batching for parallel interaction with multiple environments.

  • Evaluation of the trained agent on single episodes.

  • Video recording and visualization of the agent's gameplay.

  • Train the agent for 3000 episodes and periodically show the average agent reward during training.

    Environment: https://gymnasium.farama.org/environments/atari/kung_fu_master/

a

video.mp4

Requirements:

Python 3 PyTorch NumPy OpenAI Gym tqdm

Additional Notes:

The script currently trains 10 agents in 10 parallel environments. You can modify these numbers in the number_environments and EnvBatch class. The reward scaling (batch_rewards *= 0.01) is optional and might need adjustment depending on your environment and training dynamics.