The first release of the DDPG implementation for the Reacher Environment using Unity ML-Agents. This release includes all the necessary components to set up, train, and evaluate the reinforcement learning agent in the Reacher environment.
Main Features
- DDPG Algorithm: This release implements the Deep Deterministic Policy Gradient (DDPG) algorithm, specifically designed for continuous control tasks with high-dimensional action spaces.
- Unity ML-Agents Environment: The Reacher environment is based on Unity ML-Agents, with 20 double-jointed robotic arms learning to maintain contact with a moving target.
- Actor-Critic Neural Networks: The code includes the implementation of actor and critic neural networks, which work together to estimate the optimal policy and value functions.
- Parallel Training: The implementation leverages 20 agents collecting experiences in parallel, sharing the same actor and critic networks to improve sample efficiency and accelerate learning.
- Training and Evaluation: The repository contains a comprehensive training loop and a separate evaluation script to interact with the trained agent and environment.
How to Get Started
To get started, please follow the instructions in the README to set up the environment and install the required dependencies. Once everything is set up, you can run the training script to train the DDPG agent, and then use the evaluation script to test its performance in the Reacher environment.
I hope this release provides a solid foundation for understanding and working with the DDPG algorithm, the Reacher environment, and reinforcement learning in continuous control tasks. I encourage users to experiment with the code, provide feedback, and contribute to future improvements.
Happy learning!