Drone-AI

A project for teaching a neural network to control a drone using genetic algorithm.

Results

Great results were achieved after only 400 generations.

400_420p.mp4

This is the accompanying Youtube video: https://youtu.be/l1SYa1WjSYQ

Dependencies

• Drone interface (included)
• NN-Lib (included)
• SFML

Compiling

Linux

You can simply run compile.sh file. Compiling with this method also runs the binary.

Running

Before running make sure that /saves/[generation number]/ folders exist

Binary accepts these command line arguments:

• Learning algorithm
  • -population [size] - specify population size (default: 1000)
  • -samples [number of samples] - specify the number of reruns of each generation (default: 5)
  • -mutations [mutations / reproduction] - frequency of mutations (default: 2)
  • -max [max generation] - max generation (default: 1000)
  • -time [limit] - limit the simulation time in seconds (default: 20)
  • -load [generation number] - load a generation from saves/[generation number]/
• Graphics
  • -display - render to screen (automatically exits after one generation)
  • -width [px] - screen width in pixels
  • -height [px] - screen height in pixels
  • -record - save frames as .png to the /frames folder
  • -showctr - visualize engine input
• Other
  • -multithreading - use multiple threads for computations (only applicable when not rendering to the screen)