This repository provides a Python reimplementation of the Deep Deterministic Policy Gradient (DDPG) algorithm, a popular deep reinforcement learning method for continuous control tasks. It is designed for researchers and practitioners working with OpenAI Gym environments and TensorFlow.

How It Works

The implementation leverages TensorFlow for building and training the actor and critic networks. It follows the DDPG paper's architecture, with a key detail being the successful application of Batch Normalization to the actor network, though its implementation on the critic network is noted as problematic.

Quick Start & Requirements

• Primary install / run command:

  git clone https://github.com/songrotek/DDPG.git
  cd DDPG
  python gym_ddpg.py
  

• Prerequisites: OpenAI Gym, TensorFlow.
• To change environments, modify ENV_NAME in gym_ddpg.py. To change network architecture, adjust imports in ddpg.py.

Highlighted Details

• Implements DDPG for continuous control tasks.
• Utilizes OpenAI Gym for environment simulation.
• Batch Normalization is successfully applied to the actor network.

Maintenance & Community

No specific information on contributors, sponsorships, or community channels is provided in the README.

Licensing & Compatibility

The README does not explicitly state a license.

Limitations & Caveats

Batch Normalization on the critic network is reported as problematic. Several Mujoco environments (InvertedPendulum, InvertedDoublePendulum, Hopper) are noted as unsolved within the context of this implementation.