Measured Numbers & Honest Limits
A number without a method line and sample size is an impression. We publish our verified hardware bench numbers, simulation validation matrices, and candid physical limits in one open register.
3.3 cm
UWB Bench Std Dev
204/205 valid, 0 sentinels
0.81 ms
Formation Solve Time
137x faster than naive loop
0 / 40
Departures with Peer Link
Down from 40/40 anchor-only
0 Sats
GNSS Reliance
GPS-denied & EW-hardened
The Technical Evidence Register
All measurements below are traceable to reproducible test fixtures, bench equipment, and logged flight-session datasets.
| Metric / Capability | Measured Value | Method & Sample Size |
|---|---|---|
| UWB Ranging Precision & Yield | Mean 0.722 m, SD 3.3 cm | Continuous ~5 Hz hardware bench session, Luftschar's UWB unit N = 205 frames |
| Formation Solver Execution Time | 0.81 ms per solve | Non-linear least squares Levenberg-Marquardt on embedded Cortex-A76 N = 1,000 solves |
| Mirror Flip Rejection (Peer Links) | 0 / 40 departures (0%) | Stationary leader hover, 5-unit & 8-unit formations, 1 peer link/unit 40 independent seeds, 120 s each |
| Range Filter Ambiguity Suppression | 2 / 60 seeds flipped (3.3%) | Dynamic multi-agent formation trajectory with 8 cm Gaussian noise N = 60 random seeds |
| Filter Covariance Conditioning | 5.3 m session RMS | Kalman tracking filter with empirical covariance initialization N = 40 runs |
| True Bearing Stability Margin | 55° – 60° safe boundary | Multi-angle angular stability sweep across formation circumradius 40 seeds per bearing angle |
| Commanded Leader Yaw Ceiling | 2.0 deg/s safe rate | Grid sweep across 5 m to 60 m spacing, 40 seeds per cell N = 240 batch runs |
| Ising Ground-State Agreement | 128 / 128 planted frames (100%) | Exact mapping of follower mirror state to Ising Hamiltonian N = 128 frames at σ = 0.08 m |
| Simulator Test Suite Integrity | 40 / 40 self-tests passing | Automated headless CI test matrix 40 unit & integration tests |
Honest Engineering Limits
Publishing engineering boundaries unprompted is what gives credibility to every other claim. These are the known physics, algorithmic, and operational bounds of the current system.
Ranging alone can NEVER resolve whether the entire swarm is oriented correctly or globally reflected across the leader's anchor axis. Reflection is an isometry that preserves all mutual inter-drone distances. Breaking the global parity strictly requires an external bearing source: our calibrated optical IR camera or an off-axis sensor.
The estimator exhibits non-monotonic instability near 30° bearing: while a follower sitting exactly at 30° can be stabilized, small perturbations at 29° and 31° trigger 60/60 flip failures under anchor-only ranging. Formations must be planned outside the 25°–35° boundary or enforce peer-ranging links.
The swarm envelope is constrained by angular rate rather than linear acceleration. While airframes support 90 deg/s yaw and 4 m/s² acceleration individually, commanded leader yaw rates above 2.0 deg/s cause follower estimator divergence. Commanded trajectories must clamp turning velocity accordingly.
UWB bench precision (3.3 cm SD) was established on a static physical test stand. While RF propagation is unaffected by motor vibration, antenna phase center wobble and high-current motor EMF on companion electronics are scheduled for full in-flight validation at the BraveTechEU DefTech Forge (November 2026).
Dive Into the Technical Engineering Notes
Read deep-dive technical monographs documenting the mathematics of the mirror ambiguity, the bench testing method, and why our stack drops GNSS entirely.