Training A 2D Pixel Tank Using Reinforcement Learning

This project aims to analyse the behaviour of system of two 2D pixel tanks competing each other where one tank is trained using reinforcement learning algorithm and another tank is hardcoded (or follows a set of predefined commands). The problem is divided into 2 levels where in the first level, the tanks can only move in 1 direction and in the second level, the tanks are allowed to move freely in a 2D space. The second level is further divided on the basis of action space (discrete or continuous). Agent in each level is trained using a unique approach.

Run Main_1D.py, Main_2D_dis.py and Main_2D_cont.py to see the resluts of level 1, level 2 discrete action space and level 2 continuous action space.

Check this paper to get more insights on the project.