LogInA curious crowd simulation by Allie Bri.
Developer Allie Bri showcased an interesting crowd simulation set up in Unreal Engine, adding collision and introducing a whopping 100,000 AI agents to a bunch of boxes that create a maze of sorts.
The agents need to navigate through the obstacles, and it's clear how the majority of them are stuck at the beginning of the maze, with only a handful reaching the end.
The simulation was set up with Nanite instanced static meshes with VAT and "a lot of custom C++." The project uses less than 1,000 MB of memory, according to Bri.
Allie Bri
"The animation portion of this right now is very limited. Instance/particle space to absolute world space. Animations are changed with material custom data values based on how they were baked – start & end frame with play rate."
Allie Bri
The work features:
- Local partitioning for static & dynamic collision
- Multi-objective vector flow fields for navigation/pathfinding
- Efficient behaviors running per instance parallel on CPU.
- Nanite instancing with vertex animated textures.
- Realtime simulation at ~10ms game thread for 100K agents.
I'm sure we will see even more AI agents running and completing more complicated tasks in the future. Meanwhile, take a look at this cute AI called Albert teaching himself to walk using deep reinforcement learning.
Join our 80 Level Talent platform and our new Discord server, follow us on Instagram, Twitter, LinkedIn, Telegram, TikTok, and Threads, where we share breakdowns, the latest news, awesome artworks, and more.