I’ve recently been experimenting with AI navigation, focusing on making enemies seek out the player in a natural and intelligent way rather than just pathing straight at them. I’ve implemented multiple states — seeking, chasing, and grouping into packs — with enemies dynamically dispersing if they lose the player.

Optimising this for performance has been a challenge, especially on my non-gaming laptop without a GPU. So far, I’ve managed 1,000 enemies running these behaviours at around 7.8 FPS with a response time of roughly 200ms, which I’m very happy with. On a high-end PC, the system runs smoothly above 60 FPS.

This project has been a fun exercise in optimisation and emergent behaviour, and it’s taught me a lot about balancing AI complexity with real-time performance.

Playing ARC Raiders completely changed how I think about AI in games. Seeing how the game teaches its machine learning AI inspired me to explore similar concepts, with the goal of creating a spider-like creature with six or eight legs that can learn to walk and adapt to new terrain. This is a direction I plan to pursue after finishing my fire simulation project.

In Case you’re interested in what my ai looks like check out my code demo I did for it and gameplay footage. This is not Machine Learning Ai but instead Ai that I can say I'm proud that I made.