Drop what cannot fly, then route

Routing many aircraft is not a puzzle of joining dots. Remove regulatory airspace, terrain, weather, energy, and port occupancy first. Only then compare time, energy, and risk. This chapter is instruction. It is not a control law and not a certified planner.

Truck-and-drone models, and a pure drone net

The crowded research is the truck plus the drone. In the traveling salesman problem with a drone, and in the flying-sidekick form of it, the truck visits customers while the drone leaves the truck, serves one customer, and meets the truck again. The truck is a moving depot. It resets range. Spread across several ground vehicles, the same idea is called a vehicle routing problem with drones. Murray and Chu's 2015 formulation is one origin that later papers keep citing. The reference here is citation-level. This guide does not re-derive the equations. What it borrows is the structure: endurance cuts the sortie into a hard piece.

A pure drone network is a different problem. Depots and ports are fixed. Endurance usually forbids a chain of many landings, so a route is one sortie to one stop, or maybe two. There is a time window at the customer and another at the pad. Energy changes with wind and payload, and a battery swap takes time. Airspace capacity is not the kilogram capacity of a vehicle. The optimization looks less like a forty-stop tour and more like assigning and scheduling sorties. Work on multiple drones in the flying-sidekick setting has shown that the order of launches and recoveries is a schedule: ignore it and the endurance constraint produces plans that cannot be flown. Order is not a decoration on the geometry.

Remove the hard constraints first

What you remove first is not something you may buy your way through. Regulatory airspace: prohibited and restricted volumes, airport surfaces, temporary flight restrictions. Terrain and obstacles: the straight line on a map is not a path with vertical clearance. Weather: an arc that breaks the minima for wind, precipitation, visibility, or temperature does not exist, and it is not an expensive arc. Energy: an arc that cannot be flown with reserve, and with the return or the alternate nest, does not exist either. Port occupancy: one surface accepts one arrival or one departure in a handling interval. Encode these as high penalties and a solver under pressure will pay to cross a prohibited volume. That is a modeling error. Delete the arc, then state the objective.

The objectives come second. Time, on the customer's clock or the clinical clock. Energy, as cost and as battery life. Risk: who lives under the track, and how tangled the contingency is. Melt the three into one weight and the surprise is a fast route over a school. Keep the weights as a visible choice. A score that silently converts risk into minutes hides what was given up. In a pure drone network, unlike a truck driver waiting at the curb, a wait in the air spends energy. Do not leave 'wait' in the model as cheap slack. A wait is available only when both the energy and the pad still have room.

Conflict is a volume in time

A conflict is not a point. Two tracks may cross in plan view and still be safe if their times differ by more than the separation buffer. The converse is a shared corridor, or a single port, that meets only at an endpoint on the map and still conflicts when the times overlap. Separation is four-dimensional, buffer included. Deconfliction is a table of who is in a volume at a time, not a drawing that nudges lines apart. Solving the overlap by hovering spends the energy the previous section told you not to hide. A wait that is cheap in a truck model is not cheap for a multicopter.

Reserve the port with the same hardness as the airspace. A geometrically clean arrival that finds the surface busy does not exist as a flight. When a weather window opens and many aircraft become legal together, what jams is the slot, not the size of the sky. Solve many aircraft as simultaneous shortest paths and they will all answer with the same port at the same minute. Separating them afterward is the wrong order. Lay the slots down first and keep only the sorties that fit. Demand that does not fit is not a drone failure. It is the honest capacity of that hour.

Instruction, not a control law

What this chapter fixes is the order of decisions: delete arcs that cannot be flown, reserve port and airspace in time, and only then compare time, energy, and risk. That is a teaching model. It is not a mission file for an aircraft. Separation buffers and reserve minutes are not stated here as numbers, because they belong to the aircraft, the approval, and that day's airspace. An optimum in a paper is optimum inside the constraints the paper assumed. It is not permission for your city. Do not confuse having read the bibliographic record with having permission to fly.

Do not send this to motors. Do not feed it to the execute button on a ground station. Using this prose as a separation case, where there is no approved procedure, no surveillance that actually exists, and no operations system that can reject a plan, is unsafe. Two aircraft passing in a demonstration are not evidence that network deconfliction is in force. The person who arms an aircraft after reading this chapter, and for no other reason, is not ready. Before arming, what is required is the latest rule for that place, the flight manual of that aircraft, and an authority that is allowed to say no. The model teaches an order of thought. It does not carry responsibility for the sky. Do not upload a textbook arc.