Crate rustsim_crowd 

Microscopic crowd locomotion models for rustsim.

This crate implements the five microscopic pedestrian models catalogued at https://pedestriandynamics.org/models/ in pure 2-D continuous space plus a layered 2.5-D extension in the threed module for multi-storey environments (stations, airports, stadiums, malls).

It is called rustsim-crowd rather than rustsim-pedestrian because the same physics apply equally to cyclists, evacuees, queue-formers, and generic self-propelled agents.

Module	Model	Class	Primary references
social_force	Social Force	Force-based, 2nd order	Helbing & Molnár 1995; Helbing, Farkas & Vicsek 2000
collision_free_speed	Collision-Free Speed	Velocity-based, 1st order	Tordeux, Chraibi & Seyfried 2016
generalized_centrifugal_force	Generalized Centrifugal Force	Force-based, 2nd order	Chraibi, Seyfried & Schadschneider 2010
optimal_steps	Optimal Steps	Discrete-step utility	Seitz & Köster 2012
anticipation_velocity	Anticipation Velocity	Velocity-based, 1st order	Xu, Chraibi & Seyfried 2021
threed	Layered 2.5-D Social Force	Per-floor 2-D physics + vertical connectors	Industry standard (JuPedSim, MassMotion, Legion)

API shape

Every 2-D model shares the same three-type contract defined by the PedestrianModel trait:

• Pedestrian — the shared per-agent state (position, velocity, radius, desired speed, destination).
• WallSegment — a 2-D line segment describing a static obstacle.
• Params — a model-specific parameter bundle living inside each module.

Each model exposes:

• a Params struct with stable literature/engineering defaults.
• explicit calibration helpers for deployments that must enforce a published speed-density envelope such as Weidmann (1993).
• a unit struct implementing PedestrianModel so callers can swap models behind a trait object or generic bound.
• a free step(peds, walls, params, dt) function (deprecated since 0.0.3; O(n²) reference path retained for parity tests). Production callers use step_scratch (zero-alloc) or step_with_grid (broadphase) instead.

use rustsim_crowd::prelude::*;

let mut peds = vec![Pedestrian::new(
    [0.0, 0.0],
    [0.0, 0.0],
    0.25,
    1.34,
    [10.0, 0.0],
)];
let walls: Vec<WallSegment> = Vec::new();
let params = social_force::Params::default();
// Production hot path: zero-alloc, broadphase-accelerated.
let mut scratch = Scratch::new(recommended_cell_size(
    social_force::neighbor_cutoff(&params),
));
social_force::step_scratch(&mut peds, &walls, &params, 0.05, &mut scratch);

Design constraints

• Pure f64, SoA-friendly internals, no interior mutability.
• step_scratch(peds, walls, params, dt, &mut scratch) is the zero-alloc hot path: allocate one broadphase::Scratch per simulation and reuse it tick-after-tick. step_with_grid takes a caller-owned NeighborGrid and allocates one Vec<[f64; 2]> per tick; step is the O(n²) reference path with no broadphase.
• Determinism: every step_* variant is a deterministic function of its inputs.

Re-exports

pub use broadphase::recommended_cell_size;

pub use broadphase::NeighborGrid;

pub use broadphase::Scratch;

pub use calibration::apply_weidmann_speed_cap;

pub use calibration::apply_weidmann_speed_target;

pub use calibration::density_for_area;

pub use calibration::mean_speed;

pub use calibration::CalibrationPoint;

pub use calibration::CalibrationReport;

pub use calibration::WeidmannCurve;

pub use common::Pedestrian;

pub use common::PedestrianModel;

pub use common::WallSegment;

pub use error::CrowdError;

pub use integration::pack_columns_from;

pub use integration::step_columns_f64;

pub use integration::step_scratch_store;

pub use integration::step_scratch_store_observed;

pub use integration::unpack_columns_into;

pub use integration::AnticipationVelocityModel;

pub use integration::CollisionFreeSpeedModel;

pub use integration::CrowdAgent;

pub use integration::CrowdObserver;

pub use integration::CrowdStep;

pub use integration::GeneralizedCentrifugalForceModel;

pub use integration::OptimalStepsModel;

pub use integration::SocialForceModel;

pub use threed::LayeredObserver;

pub use threed::LayeredScratch;

pub use threed::Pedestrian3D;

pub use threed::WallPolygon3D;

Modules

anticipation_velocity

Anticipation Velocity Model (Xu, Chraibi & Seyfried 2021).

broadphase

Broadphase acceleration for crowd models.

calibration

Scientific calibration helpers for pedestrian fundamental diagrams.

collision_free_speed

Collision-Free Speed Model (Tordeux, Chraibi & Seyfried 2016).

common

Shared types and trait for all pedestrian models in this crate.

error

Parameter and runtime errors for rustsim-crowd.

generalized_centrifugal_force

Generalized Centrifugal Force Model (Chraibi, Seyfried & Schadschneider 2010).

integration

rustsim-core integration for the crowd models.

optimal_steps

Optimal Steps Model (Seitz & Köster 2012).

prelude

Convenience re-exports for the common consumer path.

social_force

Social Force Model (Helbing & Molnár 1995; Helbing, Farkas & Vicsek 2000).

threed

Layered 2.5-D crowd locomotion for multi-storey environments.