Dataset: Central and Peripheral Vision in Collective Motion

Flight recordings and analysis code for Jarvis, Palle & Floreano (2026), "Central and Peripheral Vision in Collective Motion", Laboratory of Intelligent Systems, École Polytechnique Fédérale de Lausanne (EPFL). Human collective motion is driven largely by vision, which reaches the brain through a high-acuity central region and a lower-acuity periphery. This study separates the two: participants flew inside a simulated flock in immersive virtual reality while either the central or the peripheral region of their visual field was masked, with gaze recorded throughout. The record holds every flight the paper analyses and the code that turns them into the reported figures and statistics. Experiment. Participants flew a Birdly™ (Somniacs Ltd) immersive flight simulator inside a flock of 60 birds coordinated by the Reynolds flocking algorithm, wearing a Meta Quest Pro headset that tracked gaze at 90 Hz. Each completed one 300 s flight per viewing condition, fully within-subject: Normal Vision — unrestricted field of view Peripheral Vision — central 10° radius masked Central Vision — everything outside the central 10° radius masked Normal Vision was always flown first; the two restricted conditions were counterbalanced 20/20. A further 40 flights were flown autonomously by the platform under the same flocking algorithm, giving a reference for the simulator's dynamic limits. 44 participants were recruited (mean age 24.3 ± 3.0 years; 7 female, 37 male) and 40 completed all three conditions; only those 40 appear here. Human flights were recorded 5–22 May 2025, the autonomous reference flights on 30 April 2025 and 17–18 March 2026. Contents. 160 recordings in four archives, plus the analysis code (≈2.5 GB in total): README.md — full documentation: file layout, every variable, and how to reproduce every value in the paper. Readable from this page without downloading the archives. normal_vision.zip — 40 recordings, unrestricted vision peripheral_vision.zip — 40 recordings, central 10° masked central_vision.zip — 40 recordings, outside the central 10° masked boid.zip — 40 autonomous reference flights, no human input code_and_documentation.zip — analysis scripts, figure code, generated result tables, dependency files and the licences Reproducing the analysis. Extract code_and_documentation.zip, create the environment from environment.yml (Python 3.10, numpy, pandas, scipy; matplotlib for figures), put the four condition archives in Data/ and extract them, then run python Code/01_data_analysis.py. That single script computes every metric, runs every test, writes the result tables to results/, and prints the quantities quoted in the paper's Results section. python Code/03_figures.py redraws every figure panel that is generated from data. Add --supplementary for the model-free metrics and supplementary tables. Data format. Each flight is one .pkb file: a Python pickle of a helper.flightInfo object, so helper.py (in the code archive) must be on the import path to load it. A recording holds ~2,700 samples over the 300 s analysis window (≈9 Hz) for the participant and all 60 birds — timestamps, Birdly platform pose, headset orientation basis, binocular gaze direction, platform velocity, and the 3D position and distance of every bird. Positions are in metres, angles in degrees. Gaze is sampled at 90 Hz, averaged across the two eyes and logged at ≈9 Hz. Derived quantities (flock centre, relative position, Reynolds Score) are not stored; they are recomputed on every run. Condition codes stored inside the recordings differ from the labels used in the paper. None means "no mask applied", not a missing value: None → Normal Vision BlackCircle → Peripheral Vision TransparentCircle → Central Vision Boid → Simulated (autonomous) Participants are identified only by a random four-letter code, which pairs a participant's three flights across conditions. The recordings contain no personally identifiable information; age and sex were recorded in aggregate only. Licence. Data are released under CC BY 4.0; the analysis code under the licence stated in the code archive. Funding. Swiss National Science Foundation, grant no. 200020_212077. Ethics. Approved by the Human Research Ethics Committee of EPFL, project HREC No. 092-2023. Signed consent was obtained from all participants. Citation. Jarvis B, Palle P, Floreano D. 2026 Data and code from: Central and peripheral vision in collective motion. Zenodo. doi:10.5281/zenodo.20733994 — please cite the paper alongside the record. Contact. Benjamin Jarvis, [e-mail] — Laboratory of Intelligent Systems, EPFL.