Rocket Re-enforcement Learning

I am building a RL agent that learns to land a rocket in a 3D physics sim, built from scratch. This also means that I am writing a custom rocket environment with variables such as variable mass, thrust-vectoring gimbal, finite fuel. Then I am going to train a neural net with PPO to fly powered descents and land on a moving drone ship under different gravity (Moon/Mars/Earth). The browser demo lets you watch it learn generation-by-generation with a live telemetry HUD, and races the learned agent head-to-head against a classical guidance law. This project is being made due to my interest in ML and aerospace.