❮

Evolutionary Rockets

❯

2D simulation using reinforcement learning to optimize rocket flight paths through generations of training.

1 / 4
Slide 1
DNA/flight path is completely random for the first generation
2 / 4
Slide 2
Over many generations the rockets propagate their DNA and refine the path to the target
3 / 4
Slide 3
Once the rockets have locked on to the target, they optimize the path with small mutations
4 / 4
Slide 4
The rockets will find a new path if we move the target from the last slide

Since early 2018, I've been interested in game programming, beginning with Unity-based projects. My early work includes Evolutionary Rockets, a 2D simulation that facilitated my introduction to machine learning, particularly reinforcement learning. This project features 40 rockets learning to navigate towards a target in an obstacle-filled environment, optimizing their path over successive generations.

You can try out the simulation through the webGL build provided on this website; just click the controller icon below to start.

View on GitHub Play Game