An RL agent learning to land a Falcon 9 on a droneship in simulation

An independent developer trained a reinforcement-learning agent to land a simulated Falcon 9 booster on a droneship, using only trial and error with no real flight data. The simulation models a hoverslam, in which the booster cannot hover and must reach zero speed exactly at the deck.

I trained an AI with reinforcement learning to land a Falcon 9 on a droneship, in a fully simulated world. No real flight data, just trial and error. And it’s a hoverslam: the booster can’t hover, so it has to hit zero speed exactly at the deck. 00:00