Flock behavior & herding dogs

How to Simulate

Configure the Breed and Count.
Fence: Enable to keep sheep inside the screen.
Click START to begin.
Use your Mouse to control the dog.
Herd sheep into the Brown Pen.

Status

PENNED: 0

LOST: 0

TOTAL: 30

00:00.0

Sheep Count 30

Enable Outer Fence

Boids Parameters

▼

Merino (Tight) Scottish Blackface (Loose) Custom

Separation 0.0002

Alignment 0.005

Cohesion 0.007

📊 Stats

About This Footage

▼

Aerial drone footage capturing a working shepherd dog herding approximately 800 sheep through a gate into a target pasture. This real-world observation demonstrates:

Natural flocking behavior under predator pressure
Strategic dog positioning using orbiting and weaving techniques
Herding dynamics as the flock compresses toward the gate

This real-world scenario served as the reference for calibrating our simulation parameters and validating model behavior.

What We Measured

▼

Frame-by-frame analysis extracts the flock's convex hull—the smallest polygon encompassing all sheep positions. Three key metrics quantify herding dynamics:

Hull Area (pixels²): Measures flock spread and compression under dog pressure
Hull Complexity: Number of vertices (boundary sheep) indicating flock cohesion
Centroid-to-Gate Distance: Tracks progress toward the target entrance

These geometric abstractions translate complex animal behavior into quantifiable trajectories.

Observed Patterns

▼

Analysis of the drone footage reveals consistent behavioral patterns:

Decreasing hull area: Flock compresses steadily as the dog maintains pressure
Reducing complexity: Boundary sheep decrease as stragglers join the core group
Smooth centroid trajectory: The flock center moves progressively toward the gate

Limitations

▼

This footage captures one specific herding scenario. Real-world variation includes:

Environmental factors: Terrain elevation, weather conditions, pasture layout
Animal characteristics: Sheep breed temperament, flock size, exhaustion levels
Dog expertise: Training level, breed-specific techniques, experience with this flock

Working shepherd dogs rely on visual perception with occlusion, depth estimation, and learned instinct—factors invisible to aerial observation.

How to Play?

▼

Control the dog with your mouse to herd sheep into the brown pen:

Select a breed (or customize parameters)
Adjust sheep count using the slider
Click START and begin herding

Watch how different breeds respond to pressure—tight flocks move as one unit, loose flocks scatter and regroup dynamically.

What is Flocking?

▼

Flocking is emergent behavior—no sheep "knows" the group pattern, but simple individual rules create complex collective motion.

Each sheep follows three rules:

Separation: Avoid collisions with nearby neighbors
Alignment: Match the speed and direction of neighbors
Cohesion: Stay close to the group center

That's it! These three simple rules generate all the complex patterns you see.

Sheep Behavior

▼

Merino: Tight flocking (high cohesion). Move as a dense ball, easier to guide but harder to split.

Scottish Blackface: Loose flocking (low cohesion). Spread widely, more independent, require broader dog movements.

Custom: Adjust parameters yourself to explore different behaviors!

Parameters

▼

Separation: How much sheep avoid crowding neighbors. Higher values = more personal space.

Alignment: How much sheep match their neighbors' direction and speed. Higher values = synchronized movement.

Cohesion: How strongly sheep are attracted to the group center. Higher values = tighter flocks.

Pro Tip

▼

Weave side-to-side to apply gentle pressure. Position behind the flock's center to funnel them forward. Move too fast and they panic-scatter. Too slow and they ignore you. Find the sweet spot!

Introduction

▼

Effective herding requires understanding how species-specific behaviors influence flock dynamics. This study uses Agent-Based Modeling (ABM) in the GAMA Platform to simulate shepherding.

Objective: To quantify the impact of "tight" vs. "loose" flocking behaviors on the time required to pen a group of 30 sheep.

Methods

▼

Setup: A continuous 2D environment with rigid boundaries and a target pen area.

Agents: 100 Sheep, 1 Autonomous Dog.
Logic: Extended Reynolds' Boids (Separation, Alignment, Cohesion) + Escape behavior.
Variables: Cohesion parameters were adjusted to simulate Merino (high cohesion) and Black Headed (low cohesion) breeds.

Protocol: 10 simulation runs were conducted for each breed to ensure statistical relevance.

Results

▼

Mean steps to pen (n=10):

Merino 3,996 avg steps

Black Headed 7,013 avg steps

The data reveals a stark contrast in manageability. The loose-flocking Black Headed sheep required approximately 75.5% more time to herd than the cohesive Merino breed. The standard deviation for Black Headed sheep was also significantly higher, indicating unpredictable herding events.

Discussion

▼

Interpretation: High cohesion (Merino) allows the dog to treat the flock as a single "super-agent," minimizing the need for corrective movements. Low cohesion (Black Headed) results in "splintering," forcing the dog to constantly chase stragglers.

Implication: Robotic herding algorithms must be adaptive. A "one-size-fits-all" algorithm that works for Merino sheep may fail or cause exhaustion when applied to looser breeds.

Presented as part of the seminar:
"Spatial Thinking", with topic "Spatial Simulation"

How to Simulate

Herding Performance History