- create virtual conda environment from yml file or manually
- install OpenAI GYM and PyGame using pip if needed
- install Cap-Env using python setup script (look at alias.txt file for example)
- try it out by running cap_learn.py
- design a control algorithm that generates control actions
- format of actions is a vector of 6 integers from 0 to 4 that correspond to Up, Right, Down, Left, Stay
- you may use only observations given by the environment
- you may create the algorithm in cap_learn or in a separate class/function and call it from cap_learn.
- you control one of the teams (Blue)
- UAVs (circles) cannot be killed, have large observation range and can travel over obstacles
- UGVs (squares) can kill (on their home territory) or be killed (on enemy territory) if the meet an enemy UGV
- flag can be captured by UGV only
your goals are:
- find the enemy flag
- pass through enemy patrol
- capture the enemy flag using ground units
- survive
- you may look at the Agent class if you are interested in the information beyond observations
- you are not allowed to use any information but observations from environment and our own team env.team1