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.

Understanding and Modeling Collision Avoidance Behavior for Realistic Crowd Simulation

Visually-Guided Pedestrian Crowd Simulation

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

CR-LSTM: Collision-prior Guided Social Refinement for Pedestrian Trajectory Prediction

Collision Avoidance in an Agent System for Crowd Simulation

Collision avoidance approach for example-based crowd simulation

A Collision Avoidance Model for Two-Pedestrian Groups: Considering Random Avoidance Patterns

A New Collision Avoidance Model for Pedestrian Dynamics

A Data-Driven Approach for Simulating Pedestrian Collision Avoidance in Crossroads

A Modified Heuristics-Based Model for Simulating Realistic Pedestrian Movement Behavior

Simulation of Pedestrian Behavior in the Collision-Avoidance Process Considering Their Moving Preferences

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

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

Characterization of Collision Avoidance in Pedestrian Crowds

A Hybrid Pedestrian Motion Modeling Approach for Crowd Simulation

A modified social force model considering collision avoidance based on empirical studies

Flock Based Real Time Collision Avoidance Crowd Simulation

Experimental study on pedestrians' collision avoidance

A Data Driven Simulation Model for Investigating Collision Avoidance Behavior of Pedestrians in Subway Stations

Modeling Collision Avoidance Behavior With Zero-Speed Pedestrians

Modeling Gap Seeking Behaviors for Agent-based Crowd Simulation