Crowd Simulation with Feedback Based on Locomotion State

Most crowd simulation methods treat each agent as a point and simulate the movements of these points. Although animations of virtual characters can be generated based on the trajectories of points, the generated motions often become unnatural. For example, an agent may change its walking direction suddenly in the middle of a step, even though it is impossible for humans to do so. We propose a crowd simulation method that considers the locomotion state of an agent. When an agent is in the middle of the step, a feedback force is applied to the agent to prevent the agent from suddenly changing the walking direction. However, when an agent is standing or at the beginning of the step, the agent is allowed to change its walking direction freely. We present experimental results and evaluate the effectiveness of the proposed method.