Adaptive movement behavior for real-time crowd simulation

We present a novel method for modeling adaptive movement behavior in force-based crowd simulations tailored to the requirements of interactive real-time applications. We aim for an efficient, controllable and situation-dependent synthesis of global path planning and local collision avoidance. To achieve this, we address two typical problems that can cause severe performance impacts and undesired motion: local neighbor search and the combination of goal directions with collision avoidance. We employ a direction-aware neighbor set to efficiently select agents for collision avoidance computation and crowd density estimation. We adaptively blend between goal and avoidance directions to utilize the entire spectrum of navigation strategies for goal attainment, ranging from completely goal-oriented to strongly avoidance-centric behaviors. Our method can be combined with established force-based simulation techniques and easily integrates into existing real-time simulation pipelines. Experiments show that our method can improve control over run-time performance as well as the movement behavior of agents compared to plain force-based methods.