In this environment, there are two agents whose task is to learn playing tennis. However, their goal is to learn collaborate with each other and achieve as many points as it is possible. Agent received +0.1 score for successful hitting the ball, -0.01 for letting ball hit the ground or hitting ball out of bounds. Environment is consider as solved when agents received average score of 0.5 over 100 consecutive episodes (episode score is equal to maximum of agents' scores).

The observation space of each agents is stack of three vectors corresponding to position and velocity of ball and racket in current step and two previous. Each agent can perform two continuous actions, move toward the net, and jumping.

1. Install Unity ML-agents (version 0.4) by following instruction.
2. Copy Tennis environment from Unity ML-agent directory to ./drlnd/p3_collab_compet/env. Code is compatible with single Reacher environment and multi Reacher environment.
3. Run pip install -r requirements.txt to make sure that all required python packages are installed.
4. (Optional) Add repository to PYTHONPATH: export PYTHONPATH="${PYTHONPATH}:/path/to/drlnd"

For perform learning procedure of agent just run following command in terminal:

python main.py --train

More options could be found after running:

python main.py --help

To evaluate learned policy run for example following command:

python main.py"

Description of used architecture and learning process can be found in report