Four swarm planners, one arrival-time model. Draw a field, set your camera, and compare medial axis, lawnmower, harmonic and geodesic sweeps. Every drone gets its area, path, heading and altitude, and the nadir plan is checked until 100% of the ground is imaged.
Pick a field and a camera. The planner runs in a Web Worker and streams progress for each stage.
Lime values are the best in each row. The highlighted card has the lowest max altitude at full nadir coverage.
How hard the field is to sweep: non-convexity, roughness, elongation, width variation and necks along the medial trunk.
Run a field to get a recommendation.
A Voronoi diagram of the sampled boundary gives the centres of maximal inscribed disks. Spurs are pruned by boundary arc length, and the longest path of the MST becomes the trunk. Its two ends are the start and end caps.
Each planner is a scalar field T(x): chord time along the trunk, PCA projection, the harmonic solution ∇²T = 0, or geodesic distance. The swarm formation at time t is the level set T = t.
Fronts are oriented consistently in time and parameterised by arc length s. Drone i owns s ∈ [i/N, (i+1)/N), which partitions the field. It flies the middle of its lane.
Each drone's rotated footprint 2h·tan(FOV/2) must reach every cell in its area. The footprints are stamped along every path, and any gap raises the owner's altitude until coverage is 100%.
“Harmonic fronts never fold or cusp, so they hold the lowest altitudes on bent and necked fields. Geodesic and lawnmower fronts keep workloads the most balanced.”
Asclepius / benchmark notes