Tank Battle is a multi-agent environment for deep reinforcement learning research inspired by the well-known game "Battle City 1990". The game that can be used to analyze agent behaviors and human-agent interactions in deep RL.
Tank Battle requires a complicated strategy to conquer the game. The game is in endless mode, i.e., there is an unlimited number of enemies. When the player shoots an enemy, another one will appear at the top half of the screen. The level of difficulty increases over time. The game provides the following features:
- Support different terrains
- Support multi-player setting
- Support different stages
- Written in Python, easy to extend
- Support usb controller for human player
- The game is highly stochastic