Background. Crowd simulation is widely used in fields such as urban planning, safety engineering, and interactive applications. The Boids algorithm provides a simple yet effective framework for modeling collective motion through local interaction rules. However, standard implementations typically assume homogeneous agent behavior and struggle to capture realistic dynamics in constrained environments such as bottlenecks.

Objectives. This thesis investigates how introducing heterogeneity in the steering weights of the Boids algorithm affects throughput speed and collision frequency in a constrained two-dimensional crowd simulation.

Methods. A Boids-based simulation was implemented in the Unity game engine, incorporating goal-directed movement through a waypoint system. A baseline homogeneous configuration was first established by heuristic tuning. Heterogeneous behavioral profiles (Individualist, Clinger, and Eccentric) were then created by modifying the steering weights. Each configuration was evaluated over N = 20 simulation runs with 500 agents, measuring mean throughput speed and collision frequency.

Results. The results show that heterogeneity produces distinct behavioral patterns rather than random variation. The Individualist profile reduced collision frequency but increased completion time, while the Clinger profile maintained faster throughput speed at the cost of significantly increased collisions. The Eccentric profile introduced instability, resulting in higher variability across runs. Overall, a clear trade-off was observed between throughput speed and collision frequency, with no configuration optimizing both simultaneously.

Conclusions. The study demonstrates that heterogeneous steering behaviors visibly influence crowd dynamics in constrained environments. The findings highlight the importance of accounting for behavioral diversity when modeling crowd systems and provide insight into how simple local rules can be extended to better represent varied crowd movement patterns.