NoMAC Documentation

Both models calculate in Web Workers, so you can use another tab. On HTTPS and localhost an active calculation holds a Web Lock that prevents Chrome's automatic background freezing; it is released when the calculation finishes or stops. Keep the tab open and the computer awake: other browsers and power settings may still limit background work (Chrome's freezing policy). For runs independent of the browser, use the command line.

• The supplied population file has 3,499 positive cells with B totalling 117,180. B is interpreted as a population count/weight; its census date and field definition are not supplied. Destination points are synthetic, not verified landing sites.
• Grid spacing is independent of the density-map resolution. Layers are 40 m apart by default; protective cylinders of ±15 m leave a nominal 10 m gap, and layer spacing must exceed every aircraft's protection height. These are experiment assumptions, not validated separation minima.
• One request is one continuous round trip. Straight junctions are passed without stopping; turns and vertical transitions respect the aircraft's horizontal and vertical acceleration limits. Restrictions and conflicts are checked along every phase.
• Delivery hover defaults to 30 s at 10 m above flat reference ground and only counts within 1 m of the delivery point. Missed or interrupted service is reported separately. Takeoff/landing contact, ground service, batteries and fleet size are not modelled.
• Geofences change routes, waiting and rejection; trajectory mode does not redistribute Dn mass. Dn is not a demand target for daily scenarios; older imported Dn-calibrated projects keep their original demand model.
• A day is limited to 2,000 flights. Demand whose expected count plus five standard deviations would exceed that limit is rejected before running.

Equations and requirement mapping: daily traffic contract and delivery model contract.

• Each aircraft can use its own sensor profile. FoV is a horizontal bearing cone around the heading; no vertical FoV is modelled. Position error is bounded per axis and velocity is exact. CA predicts the intruder at constant velocity from observations, never from its future path.
• All DAA profiles share one reference vertical avoidance policy; they are parameter sets, not independent supplier algorithms.
• False alarms, first-failure hazards, smooth vertical tracking deviations and urgent geofences each use a separate seeded random process.
• After a failure the aircraft brakes and hovers with an enlarged spherical protection volume, then attempts a bounded rejoin.
• A collision-first geofence policy may relax a restriction to avoid predicted contact. Physical solids and the airspace boundary are never relaxed; policy use and violations are recorded.
• Groups of three or more and overlapping avoidance maneuvers are flagged, not solved.

• Completed flights per demand hour describe the offered cohort, including completions in the finish period. They are not steady-state throughput.
• If the highest tested level passes, it is a lower bound; no maximum has been located. A result applies to the chosen traffic and planning policy, not to airspace in general.
• Inspect run reproduces any underlying run with its exact seed and settings.
• The reference covers one flat-ground area and multirotors. See the traffic model contract. Engine fixtures for scripts/traffic.cjs: corridor network, contingency, imperfect sensing, non-cooperative intruder.

• Each motion segment has constant acceleration. Multirotors accelerate and brake on each leg within separate horizontal and vertical acceleration limits, stop at turns and take time to yaw. Heights are AGL above flat reference ground.
• Event checks solve the continuous relative-distance and polygon-prism geometry, so changing the trace interval changes the display only. In the top view, overlapping circles alone do not establish a 3D conflict.
• The CA reference tries an upward, then a downward offset that keeps the predicted protection volumes apart. If overlap has already begun, it may instead restore separation sooner, without new NMAC or contact. It then returns continuously to the original timed route.
• The two-aircraft reference experiment is available from the command line: node scripts/trajectory.cjs --example crossing (add --daa-off for the baseline). It does not estimate capacity, validate TLS or represent supplier DAA performance. See the trajectory model contract.

• Batches keep aggregate counts, exposure, occupancy and up to 12 reproducible seeds of the most eventful days, not every flight path.
• The event list keeps the first 400 events per type for replay; totals always include every event and truncation is shown. Selecting an event regenerates its day and opens its hour, paused at the event with the aircraft selected.
• Stop keeps the completed days, and exported partial batches can be resumed. A flight-hour target finishes whole days, so exposure may slightly exceed it.
• Millions of flight hours are a requested stopping target, not a demonstrated capability; full-day studies are slow in the browser.

Shared engine, file formats and commands: integrated model contract.

• Dn is the relative traffic weight in the GeoJSON file, not a real-time probability that a drone is there. The map is a bounded analysis area; the tool does not invent Dn outside or between source cells.
• Overlay colours are contrast-enhanced and clipped near the top, so colour differences are relative. Satellite imagery (ArcGIS World Imagery) does not enter any calculation.
• The active-drone count per snapshot is Poisson-sampled. Sampling is seeded, so every radius is evaluated on the same traffic and an unchanged scenario reproduces its result.
• Flight hours/day = operations/day · flightMinutes / 60. These are scenario metadata; the snapshots have no physical-event exposure.

Source: LiU probabilistic traffic map, QRA demand-map approach and the local GeoJSON Dn field. Display scaling is an implementation choice.

RWY5K is a bundled local geometry preset, not an authoritative airspace restriction.

Source: product description for geofences and height restrictions; the QRA paper and AST notes motivate boundary congestion. The edge-push algorithm and its parameters are project sensitivity choices.

The blue disk is a mean-centred implementation envelope, not a minimum enclosing circle. It touches the outer edge of the furthest red safety circle and contains the others, so there can be extra space on other sides; tightening it would change A and possibly the classification.

The free area subtracts the union of geofences inside the disk once, using analytic polygon-edge and circle-arc integrals. Red safety disks are safety margins, not blocked airspace, and are not subtracted. The map edge is not a wall. In the inspector, green and red dots only illustrate free and blocked regions; large dark dots are drone centres.

The Java prototype merges disks recursively. This tool first finds the whole connected group, so radii are not simply added as 3r, 5r and so on.

Source-backed concept: LiU conflict graph/connected components and ATM overlapping safety disks. Mean-centred envelope: project implementation choice.

A/n can hide a dense subgroup: two coincident drones at r = 50 m are classified, but a third drone 100 m away enlarges the blue disk enough to remove the classification, although the coincident pair is still there. Pair counts and the closest-pair distance are therefore shown separately in the inspector.

A former 3D surface encoded the same 0/1 classification as height and was removed. The 2D map's design follows the supplied “sharp 0/1 boundary” reference figure. Legacy comparison targets in older files are only kept for import compatibility. See corrections and validation for analytical checks.

Background positions are held for inspection and replaced every five simulated minutes; they do not represent continuous drone trajectories. U-plans move continuously along their planned paths and are evaluated separately. At 60:00 the 55:00 sample stays visible; there is no thirteenth sample. The Airborne badge counts sampled background drones only.

The status names the scenario hour and the clock shows elapsed time within it. Earlier random draws are consumed when a later hour is inspected, so its samples stay aligned with the full run. The view shows the A/n research model, not executed avoidance or physical collisions; aggregate statistics remain in Results. See the interface review.

• Conflict intervals are merged per pair; a pair that separates and conflicts again counts twice. Fast crossings between display samples are not missed.
• Reroute all handles plans one at a time, so the result is heuristic and depends on order. A start inside an active geofence band is reported as an error.
• Generated routes avoid geofence bands where possible but not each other until rerouted.
• The conflict list shows the pair, interval start/end and minimum separation. Selecting a closest approach moves the planned-flight preview to that instant. Playback follows constant-speed plans and is not an avoidance simulation. Long lists are paginated; JSON export keeps every interval.
• Original plans survive edits, reroutes and export/import; older files without one use the imported plan as the baseline. Ratios with a zero denominator are exported as null, and a shorter route can exceed 100% efficiency.
• Result tables: pair intervals count once, the per-route table counts each interval for both routes, and geofence intersections count each route–geofence pair once. Closest flagged approach is the smallest horizontal separation among detected conflicts.
• Presets: Add RWY5K adds the bundled restriction; Add crossing U-plans adds two equal, perpendicular routes that meet at the same altitude and time. Both keep the existing settings and routes.

Source: product description for U-plans, randomized flights and {course, distance, height} legs; QRA route-GUI inspiration. CPA and A* routing are implementation choices.

• Metrics: A/n classifications per snapshot, A/n observations per day, groups with 3+ drones per day, and the largest group. Bars share a linear scale; Δ is the change from the reference run at the same radius (absolute when the reference is zero). Differences are descriptive, not significance tests.
• Runs & scenario settings lets you rename or remove runs, inspect provenance, or Open in Scenario, which restores the saved geometry and recalculates with the current engine.
• Export all saves the comparison as a reusable batch with the reference first. Limits: 30 runs and 10 MB per file. The comparison lasts until the page is reloaded.
• The dataset is identified by SHA-256 on HTTPS or localhost. Older exports remain readable, and unrecorded metadata shows as Not recorded. An incomplete run can be exported but not compared as evidence of zero events.
• Trajectory projects export their scenario, traffic sources, profiles, seeds and study results, and can be resumed or replayed after import.