How do you move from a handful of NPCs to an army of thousands? This talk explores building a real-time crowd simulation from scratch, stripping away “heavy” NPC logic for lean, data-oriented agents. We’ll dive into the most popular movement models, from Boids and ORCA to Social Forces, and compare pathfinding schemes like navigation graphs and flow fields. Learn how to scale your simulation, all illustrated with visual demos.
1. “Regular” game NPCs are too heavy for crowds, a simplified agent representation is needed.
2. Different crowd densities require different simulation models: local models for low densities, global models for high densities.
3. Efficient pathfinding requires going beyond standard navmesh queries for individual agents: graphs and flow fields are more effective.
4. Maintaining performance means embracing data oriented programming and leveraging CPU multithreading or GPU offloading.
