Velocity-based dynamic crowd simulation by data-driven optimization

A crowd simulator that generates realistic crowds with various movement patterns and environmental adaptability is urgently desired but underdeveloped for the applications of video games, urban visualization, autonomous driving, and robot navigation test. In this work, we present a novel velocity-based framework based on data-driven optimization to build dynamic crowd simulation that allows interactive control of global navigation, local collision avoidance, and group formation. An agent's adaptive decision-making regarding its goals and dynamic local environment is formulated as an optimization problem which is solved by finding an optimal velocity from the real-world crowd velocity dataset. Each component that affects an agent's movement is integrated into a velocity-based crowd energy metric to measure the similarity between the agent's required simulated velocity and a given velocity sample. The proposed model can simulate thousands of agents at interactive rates. In addition, the framework is general and scalable to be integrated with various crowd simulation methods to meet the requirements of various kinds of robot navigation test. We validate our approach through simulation experiments in robot navigation scenarios, as well as comparisons to real-world crowd data and popular crowd simulation methods.

Automated summary

The study introduces a novel velocity-based framework for dynamic crowd simulation, offering interactive control over crowd movements, simulating thousands of agents at interactive rates, and being both general and scalable for various robot navigation tests. Validation is performed through simulation experiments and comparisons to real-world data and existing crowd simulation methods.