Helicopter Game

This is a helicopter game adapted from PyGame Learning Environment's Pixelcopter, available at https://github.com/ntasfi/PyGame-Learning-Environment

This project is for Intelligent Systems at Mälardalens Högskola. It involves several learning algorithms that learn how to play the game.

See NEAT playing the helicopter game here

Clipart for the helicopter is taken from https://www.clipartmax.com/download/m2i8K9i8b1A0b1N4_free-to-use-public-domain-transportation-clip-art-helicopter-clipart/

---

Prerequisites:

• argparse
• matplotlib
• numpy
• pygame
• graphviz

Included Libraries:

• pygamewrapper
• vec2d

Usage:

For a human player:

./pixelcopter.py

To control the simulation speed (frames per second):

./pixelcopter.py --fps 999

The default value is 30 fps.

For an AI agent player whose class is stored in the agents/ subdirectory and called stupid_agent.py:

./pixelcopter.py --agent stupid_agent

To run the game using noisy positioning data:

./pixelcopter.py --agent stupid_agent --noisy-sensors

To run the game without animation (--quiet-mode), for quicker testing of an agent:

./pixelcopter.py --fps 4000 --agent stupid_agent --quiet-mode --headless

By default, the game is in 'flappy' mode, like in Flappy Bird. However, it can be run in 'helicopter' mode:

./pixelcopter.py --mode helicopter

Agent Structure

Agents must be contained in a uniquely-named .py file within a class Agent. Agent selection and import is done at the terminal at runtime as described above. Agents must contain the following functions:

• Agent(self) is the constructor called by the game.
• get_action(self, game_state) is called by the game once per frame. It supplies the agent with the current state of the game and expects an action.
• reset(self, game_state) is called by the game in the event of a game-over.

Agent Inputs

• normalized_heights: array of normalized values representing the helicopter's closeness to the top and bottom of each of the next 3 terrain sections.
• sensors: 5 proximity sensors aimed in front of the helicopter. 0 means no obstacle. Anything above 0 represents an obstacle at the height of the sensor, and increases to 1 as the obstacle becomes closer.
• y_velocity: how quickly the helicopter is moving upwards or downwards. Not normalized.