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.

Less-Effort Collision Avoidance in Virtual Pedestrian Simulation

Inserting Virtual Pedestrians into Pedestrian Groups Video with Behavior Consistency

Avoidance Behaviors of Pedestrians in a Virtual-Reality-based Experiment

Modeling Collision Avoidance Behavior With Zero-Speed Pedestrians

Understanding and Modeling Collision Avoidance Behavior for Realistic Crowd Simulation

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

Real-Time Navigation In Dynamic Human Environments Using Optimal Reciprocal Collision Avoidance

A Modified Social Force Model Considering Relative Velocity of Pedestrians

Improved Social Force Model based on Pedestrian Collision Avoidance Behavior in Counterflow

The Improved Velocity-based Models for Pedestrian Dynamics

Pedestrian Interaction Algorithm for Autonomous Vehicles Considering Crossing Intention and Driving Style

Development of pedestrian collision avoidance strategy based on the fusion of Markov and social force models

Generalized Collision-Free Velocity Model for Pedestrian Dynamics

Robot Navigation in Dynamic Human Environments Based on Optimal Reciprocal Collision Avoidance

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

Coupling intention and actions of vehicle–pedestrian interaction: A virtual reality experiment study

A Local Behavior Model for Small Pedestrian Groups

DeepORCA: Realistic crowd simulation for varying scenes

Kinetic and Kinematic Features of Pedestrian Avoidance Behavior in Motor Vehicle Conflicts

A collision-free model on the interaction between pedestrians and cyclists on a shared road

Pedestrian Evacuation Modelling with Dynamics Congestion Avoidance