Abstract:For walking pedestrians, when they are blocked by obstacles or other pedestrians, they adjust their speeds and directions to avoid colliding with them, which is called collision avoidance behavior. This behavior is the most complex part of pedestrians' walking processes and its modeling and simulation are the keys to realistic crowd simulation, which serves as the foundation for various applications. However, most existing methods either lack the representation power to accurately model the complex collision behavior or do not model it explicitly, which leads to a poor level of realism of the simulation. To realize realistic crowd simulation, we propose to analyze, understand, and model the collision avoidance behavior in a data-driven way. First, to automatically detect collision avoidance behavior for further analysis, we propose a domain transformation algorithm that detects it by transforming the trajectories in the spatial domain into a new domain where the behavior is much more apparent and is thus easier to detect. The new domain also provides a new perspective for understanding collision avoidance behavior. Second, since there are no mature metrics to evaluate the level of realism, we propose a new evaluation metric based on the least-effort theory, which evaluates the realism of collision avoidance behavior by its physical and mental consumption. This evaluation metric also provides the foundation of modeling. Third, for realistic crowd simulation, we design a reinforcement learning model. It trains agents with our proposed reward function that models pedestrians' intrinsic needs of "reducing effort consumption'' and thus can guide agents to behave realistically when avoiding collisions. Extensive experiments show our model is 55.9% and 52.5% more realistic in collision avoidance behavior than the best baselines on two real-world datasets. We release our codes at https://github.com/tsinghua-fib-lab/TECRL.

A Generic Bi-Layer Data-Driven Crowd Behaviors Modeling Approach

Learning Behavior Patterns from Video for Agent-Based Crowd Modeling and Simulation

Visually-Guided Pedestrian Crowd Simulation

Learning Behavior Patterns from Video: A Data-driven Framework for Agent-based Crowd Modeling.

Model-based Crowd Behaviours in Human-solution Space

A Role-dependent Data-driven Approach for High Density Crowd Behavior Modeling

Understanding and Modeling Collision Avoidance Behavior for Realistic Crowd Simulation

Modeling and realizing methods for virtual crowd behaviors

Agent-based Human Behavior Modeling for Crowd Simulation

Simple Cognitive Heuristics Applied to Modeling Pedestrian Behavior Dynamics

A collective motion model based on two-layer relationship mechanism for bi-direction pedestrian flow simulation

Modeling Dynamic Groups for Agent-Based Pedestrian Crowd Simulations

Data-Driven and Collision-Free Hybrid Crowd Simulation Model for Real Scenario

A Force-Based Model for Adaptively Controlling the Spatial Configuration of Pedestrian Subgroups at Non-Extreme Densities

A Cellular Automata Based Crowd Behavior Model

Simulating Realistic Crowd Based on Agent Trajectories

Modeling, Simulation and Visual Analysis of Crowds: A Multidisciplinary Perspective

Crowd Modeling and Simulation Technologies.

Diverse Motions and Responses in Crowd Simulation

Understanding complex crowd dynamics with generative neural simulators

Effects of Crowding Perception on Self-organized Pedestrian Flows Using Adaptive Agent-based Model