Vision-based-Reinforcement-learning-control-for-aircraft

This is a project using DQN-PER and video frame to train an(or more) intelligent aircraft(s) from scratch.

Features:

1. Multiprocess
2. Two types of communication: UDP or DDS
3. Vision based (or data based or mixed) reinforcement learning control for aircraft(s)

we provide two tools you can use in our project:

1. Vision based aircraft analysis (1. hit prob 2. frame stack)
2. DDS communication for multi-aircrafts (NOTE: You need compile OpenDDS in your local environment. Folders dds_test6 and dds_test7 are our test demo that you can refer. Also, we provide a powershell script to open 4 cmd windows transferring packages.)

You can find our demo video here.

Disclaimer

We love peace, and all of our offensive words such as air combat are limited to simulation games, so please do not misinterpret our work. We do not accept any accusation of militarization.