In this project, I will solve a continuous control problem using deep reinforcement learning. I will work with the Reacher environment, from Unity ML Toolkit, that contains 20 identical agents, each with its own copy of the environment. The goal of the agent is to maintain its position at the target location for as many time steps as possible, as demonstrated below:
A reward of +0.1 is provided for each step that the agent's hand is in the goal location. The observation space consists of 33 variables corresponding to position, rotation, velocity, and angular velocities of the arm. Each action is a vector with four numbers, corresponding to torque applicable to two joints. Every entry in the action vector should be a number between -1 and 1.
To solve the environment, the agents must get an average score of +30 over 100 consecutive episodes, and over all agents. This project is part of the Deep Reinforcement Learning Nanodegree program, from Udacity. You can check my report here.
This project requires Python 3.5 or higher, the Reacher Collector Environment (follow the instructions to download here) and the following Python libraries installed:
In a terminal or command window, navigate to the top-level project directory reacher-rl/ (that contains this README) and run the following command:
$ jupyter notebookThis will open the Jupyter Notebook software and notebook in your browser which you can use to explore and reproduce the experiments that have been run.
- Lillicrap, T. P., Hunt, J. J., Pritzel, A., Heess, N., Erez, T., Tassa, Y., et al. Continuous control with deep reinforcement learning. arXiv.org, 2015.
- Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., et al. Human-level control through deep reinforcement learning. Nature, 2015.
- ...
The contents of this repository are covered under the MIT License.