Data and code from: The aerial combat strategy of dragonflies

Fabian, Sam 1 ; Yarger, Alexandra2; Chen, Szu-Lung3; Lin, Huai-Ti2

Published May 01, 2026 on Dryad. https://doi.org/10.5061/dryad.qrfj6q5ws

Data files

May 01, 2026 version files 10.06 MB

Dragonfly_Data_and_Code.zip

10.05 MB
README_for_Dragonfly_Aerial_Combat.md

6.90 KB
README.md

6.80 KB

Abstract

Dogfights are aggressive aerial contests between evenly matched competitors. Male dragonflies engage in such interactions to establish territorial dominance. Using field stereographic recordings of Trithemis aurora, we examined dragonfly behavioural objectives during these encounters. Unlike predatory pursuits, in which intercept trajectories minimise time-to-contact, male T. aurora steer to keep their opponent forward and elevated within their frontal field, modulating speed to avoid direct collision. These contests feature frequent role reversals, with evenly matched rivals alternating between chaser and evader. The manoeuvres observed during these exchanges, including looping and spiralling flight, emerge from the underlying pursuit objectives. During territorial conflicts, males exhibit exceptional agility with centripetal accelerations of up to 6 g. Despite this high performance, individuals spent approximately one-third of flight time gliding in short bursts, even during close combat. Our findings show that complex aerial contests can arise from simple pursuit control principles constrained by sensorimotor limits.

Associated Publication

Samuel T. Fabian, Alexandra M. Yarger, Szu-Lung Chen, Huai-Ti Lin; The Aerial Combat Strategy of Dragonflies. J Roy Soc Interface, 2026

Dataset Overview

This dataset contains 3D reconstructed flight trajectories of aggressive aerial interactions between male dragonflies (Trithemis aurora), along with MATLAB scripts used for kinematic analysis and simulation studies presented in the associated publication. All data and code is located in the file titled 'Dragonfly_Data_and_Code.zip'. This readme is also enclosed as 'README_for_Dragonfly_Aerial_Combat.md'.

The dataset includes:

Reconstructed 3D trajectories from stereo high-speed video recordings
Paired interactions (dogfights) between two male dragonflies
MATLAB scripts for:
- Kinematic analysis and figure reproduction
- Simulation of pursuit-based guidance strategies

Dataset scale:

102 paired interactions
204 individual trajectories
Median trajectory length: ~1.06 s
Recordings captured at 250 fps, typically with ~2000 frames per recording

Each recording may contain multiple interactions, but each interaction involves only two individuals.

A trajectory corresponds to a single continuous track of an individual dragonfly during an interaction.

File Structure

All Dogfighting Trajectories/

Contains .mat files for individual recordings. Each file may include one or more interactions between two dragonflies.

functions/

MATLAB helper functions for data loading, kinematic calculations, and plotting. Automatically added to the MATLAB path by the main scripts.

main_DogfightKinematicAnalysis.m

Primary script used to reproduce analyses and figures from the publication.

main_SteeringDuel.m

Simulation of two interacting agents using pursuit-based control laws. Multiple configurations are available in the script comments.

main_SingleRecordingPlotting.m

Example script for visualising trajectories from a single recording.

Experimental Methods Summary

Recordings were made using a calibrated stereo camera system with frame-synchronised high-speed cameras.
Videos were recorded at 250 frames per second, typically with ~2000 frames per recording.
The recording volume was approximately 2 m x 2 m x 1.5 m around freshwater ponds.
2D trajectories were manually curated and triangulated to reconstruct 3D motion.
Trajectories were smoothed using cubic spline fitting after removal of high-error points.
Each interaction consists of two male dragonflies engaged in aggressive aerial pursuit, with frequent role switching between chaser and evader.

Data Structure

Each file named DF_xxxx_3D_Trajectories.mat contains two MATLAB structures:

1. metadata

Variable	Description	Units	Notes
stereoParams	Stereo camera calibration parameters	-	Used for triangulation
fps	Frame rate	Hz	Typically 250 Hz
smoothCnst	Smoothing constant	-	Used in spline smoothing
smoothDistLim	Distance threshold for filtering	mm	Excludes outliers
smoothRMSEThresh	Reprojection error threshold	pixels	Excludes poor reconstructions
trackC1, trackC2	2D tracked positions	pixels	Cell arrays of [X,Y]; NaN = not visible
nf	Number of frames	-	Typically ~2000

2. DF_interactionData

Variable	Description	Units	Notes
rawTraj	Raw 3D trajectories	mm	Cell array of [X,Y,Z]; cell 1 is often empty in dyadic interactions
rawTrajRMSE	Reprojection error	pixels	Per-point reconstruction error
smoothTraj	Smoothed 3D trajectories	mm	Filtered trajectories used for kinematic analysis
flappingState	Wing state classification	-	1 = gliding, 2 = flapping

Coordinate system: positions are expressed relative to camera 1 and rotated so that the Z-axis is aligned vertically using the plane of the water surface.

Reference frame: trajectories are expressed in the ground reference frame. Wind effects are assumed negligible because recordings were taken in a sheltered location.

Behavioural Context

All trajectories represent male-male aerial contests (dogfights).

Key behavioural features:

Interactions involve two individuals with no physical contact
Roles (chaser and evader) frequently switch during an interaction
Flight paths include looping, spiralling, and tight turning manoeuvres
The behavioural objective is not interception, but maintaining a favourable relative position

How to Use This Dataset

Load a trajectory file in MATLAB:
```
load('DF_xxxx_3D_Trajectories.mat')
```
Access trajectory data:
```
DF_interactionData.rawTraj
```
Key usage notes:
- rawTraj contains reconstructed 3D positions
- In most files, rawTraj{1} is empty
- Remaining cells correspond to individual dragonflies
- Units are millimetres (mm)
For example visualisation:
```
main_SingleRecordingPlotting.m
```

Data Usage Notes and Limitations

Sections of missing data are represented as [NaN, NaN] in 2D tracks.
Reconstruction accuracy varies; points with high reprojection error (>5 pixels) may be unreliable.
Smoothed trajectories are provided for convenience but may not be suitable for all analyses.
Users performing detailed kinematic analysis are encouraged to reprocess raw trajectories as needed.
Individual dragonflies are not tracked across recordings; trajectories represent independent interaction segments.

Reproducing Analyses

Main kinematic analyses:
```
main_DogfightKinematicAnalysis.m
```
Simulation of pursuit interactions:
```
main_SteeringDuel.m
```

Requirements

MATLAB R2022b (or compatible version)

Contact

Samuel T Fabian,

University of Oxford,

samueltfabian@gmail.com