Algorithms for Microscopic Crowd Simulation: Advancements in the 2010s

van Toll, Wouter; Pettré, Julien · 2021 · OpenAlex-citations

DOI: 10.1111/cgf.142664

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This state-of-the-art report analyzes advancements in microscopic crowd simulation algorithms from 2010 to 2020, focusing primarily on local navigation and collision avoidance. The authors address the need for a critical review of this rapidly growing field, which models crowds as individual agents whose collective behavior emerges from local interactions. While previous surveys provided broad overviews, this work specifically examines the evolution of algorithmic categories within the microscopic paradigm, aiming to identify conceptual differences, advantages, and future research directions. The motivation stems from the increasing volume of publications in computer graphics and robotics, with Google Scholar data indicating a surge in research activity, peaking at nearly 700 publications per year by 2019. The study categorizes local navigation algorithms into four main types: force-based, velocity-based, vision-based, and data-driven methods. The authors define the standard simulation loop, where agents update their velocities based on neighbor information and preferred velocities derived from global path planning. They detail collision-prediction concepts such as time to collision, distance to closest approach, and bearing angles, which underpin many modern avoidance strategies. The review filters literature to include only papers introducing new concepts or algorithms, excluding mere applications of existing methods. It also covers related topics such as group behavior, queue following, and the integration of local navigation with global path planning. Key findings highlight the maturation of velocity-based algorithms and the emergence of vision-based methods that simulate human perception, as well as significant growth in data-driven approaches leveraging deep learning. The authors analyze the computational and conceptual trade-offs of each category, noting that while microscopic methods are preferred when individual differences matter, macroscopic methods remain useful for dense crowds. The report identifies a clear need for better evaluation metrics to quantify simulation realism and guide algorithm selection. It concludes that future research will likely see increased adoption of learning-based navigation and methods that unify multiple behavioral levels into single principles. Additionally, the authors predict a growing emphasis on analyzing real and simulated crowd behavior to improve the accuracy and applicability of these simulation tools.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• traffic density