Implementation of a Deep Reinforcement Learning agent to solve the Unity Reacher environment.
This repository contains all the material relative to the implementation of the Project 2 of the Udacity Deep Reinforcement Learning Nanodegree program.
The environment consists of a double jointed arms that can move to a target location. The goal of the agent is to maintain the position at the target location for as many time steps as possible. The full environment consists of 20 identical copies of the double jointed arm.
The following description is for a single double-jointed arm.
-
Size: 33
-
Content: Position, Rotation, Velocity and Angular Velocities of the arm
-
Size: 4
-
Actions: Torque to be applied to the joints
- +0.1 : for each step that the agent's hand is in the target location
The environment is considered solved when the agent gets an average score of +30 over 100 consecutive episodes, and over all agents
Follow the instructions below to install the software necessary to run the agent in the environment.
1. Set up the Python environment on your machine by following the instructions in the DRLND GitHub repository.
2. Download the Unity environment:
3. Place the unzipped folder just downloaded in the p2_continuous-control/ folder in the DRLND GitHub repository.
4. Clone this repository in the p2_continuous-control/ folder in the DRLND GitHub repository.
The agent can be tested and trained in the Continuous_Control.ipynp python notebook. Following the instructions in the notebook it is possible to see our trained agent in action or to train a new agent using the hyperparameters of your choice and to see your agent in action.