I have started my Stardance project which is teaching a neural network to land a rocket in 3D. My plan is to build a custom rocket physics sim from scratch which includes variables such as variable mass, thrust-vectoring gimbal, fuel limits. Then train an RL agent to fly powered descents, and put it in the browser so you can watch it learn. Got the Git and Python/TS toolchains set up today. Next, I am working on the physics core