SWARMSCOPE

Distributed MPC · MHE — cluster flight replay
GNSS BLACKOUT
T+00.0 s
DMHE err
—
DR drift
—
Min sep
—

Denied-agent position error

Inter-agent separation

Formation error

SWARMSCOPE

A replay of real, validated closed-loop simulation data

Every trajectory, attitude quaternion, estimate and metric on this page is raw logged data from the validated distributed estimation-and-control study: six hexacopters (full 13-state quaternion plants) fly take-off, ring assembly, a swept translation and a contract-and-rotate reconfiguration at 20 Hz, coordinated by distributed MPC with consensus and a control-barrier-function safety filter. Nothing is key-framed.

At T+12 s agents 3·4·5 lose GNSS for ten seconds. Watch the red wireframe ghosts — the dead-reckoning estimate — drift more than a metre while the distributed moving-horizon estimator fuses relative measurements to GNSS-good neighbours (cyan beams) and pins each denied agent to within ≈9 cm, keeping the formation and the 1.4 m safety margin intact.

Reading the scene
Coloured airframesagents 0–5; LED ring = agent colour, amber pulse = GNSS denied
Glowing rings on the ground planecommanded formation slots (consensus centre + offsets)
Faint blue linkscommunication graph (ring + 2-hop chords)
Cyan dashed beamsrelative-measurement fusion to GNSS-good anchors
Red wireframe + tetherdead-reckoning estimate and its drift from truth
Amber pair lineclosest pair; safety radius D = 1.4 m
Headline numbers (Monte-Carlo, from the study)

denied-agent RMSE  DMHE 0.09 m · dead-reckoning 0.96 m
min separation ≥ 1.55 m · margin breaches 0 · divergences 0/210 runs
embedded solvers in-loop ≈ 76 µs · agent · cycle

Fixed-wing scene: four 6-DOF aircraft on a coordinated echelon loiter (R = 180 m, Va = 25 m/s, h = 120 m). Airframes are drawn ≈2× scale and rotor/propeller spin is slowed for visibility; all poses come from the logged states. Drag to orbit · wheel to zoom · right-drag to pan. — Sreeram Anil

← Project