top of page

SPACE AGENTS

2023 - In this project I implemented Unity's ML Agents plugin to see if reinforcement learning can be used on satellites to learn rendezvous without any knowledge of gravity. Observation and rewards were based only on relative velocity and position. One continuous action space was used to apply force along the in-track axis. Training took about 30 minutes to plateau to a low fidelity min time rendezvous. Checkout the images below for training shots and final modeling rendering.

bottom of page