How the AI works
Twenty landers evolve with NEAT under a simple physics model (gravity + thrust). Each network must fire engines to land softly on the platform.
Inputs → outputs
- Inputs: position, velocity, angle, and distance to the pad.
- Outputs: main thrust and left/right thrusters.
Fitness
Reward depends on landing softly and on target; crashing or drifting away is penalized — so evolution favors gentle, accurate descents.
What you see on screen
A value heatmap highlights the safe zones the population has learned, and successful soft landings become more frequent each generation.