Automated livestock herding

Move the mob without the quad and the gate

A drone walks a mob from one grazing block to another on a route you approved, inside limits you set, and stops if it cannot do it calmly.

Watch it work on the playground, where the same controller that would fly the aircraft is run against a modelled paddock.

The problem

Every rotational grazier's day has a move in it: fresh cover, a gate, a mob that would rather stay. It is twenty minutes and two people on a good day, and on a bad one it is an animal on a road at six in the morning after a fence has gone down. The blocks furthest from the yard get moved late for the same reason they get checked least.

How it works

What you get

• An approved route per move
• Head count on arrival
• A straggler list, so somebody walks out to whoever stayed
• A welfare record per move
• A grazing rotation log

What the dashboard measures

• Closest the aircraft came, against its standoff
• Fastest the mob moved, against its limit
• Minutes under pressure
• Animals left behind per move
• Moves stopped, and why

Where it matters most

Dairy and grazing, grassland, cooperatives.

Build it

This is the only capability in the catalog that acts on live animals, and that one fact decides the whole build. Everything else here scans ground or treats it. Get a weed map wrong and you spray four hectares you did not need to. Get this wrong and you have run somebody's in-calf heifers through a fence.

Welfare law is the specification, not the caveat

In Germany §3 of the Tierschutzgesetz forbids driving an animal beyond its capacity or causing it avoidable distress. Equivalent duties exist across the EU and in the UK. "The software was careful" is not an answer to that, so the limits have to be named, agreed to, recorded and answerable:

The envelope is copied onto the move record when a person authorises it, rather than read from configuration when the aircraft launches. A settings change next month must not be able to rewrite what somebody agreed to, and an audit a year from now has to read the same figures they saw.

Three phases, because a cron entry must not be able to fly this

A herding move is planned automatically, authorised by a person, and only then flown, inside a window. The nightly job can propose and nothing else. There is no code path from the scheduler to an aircraft over livestock.

An authorisation lapses if nothing acts on it. Somebody agreeing to a drone working their stock at seven in the morning has not agreed to it happening at dusk.

The controller is stockmanship, not path planning

Two ideas do nearly all the work, and both are older than drones.

The flight zone is the radius inside which an animal moves away from you and outside which it ignores you. Pressure is applied by entering it and released by leaving. The point of balance is why the pressure has to come from the side you want them to leave: stock move away from pressure applied behind the shoulder, so the aircraft belongs behind the mob relative to where it is going, never above it and never in front.

Three consequences a naive implementation gets wrong:

• Pushing harder does not make a mob move better. Past a point it makes them run, and a running mob splits. The correct response to a mob that has speeded up is to back off, which reads backwards until you have watched it happen.
• Height is the better pressure release. Backing away horizontally gives up the point of balance and the mob stops walking. Climbing sheds pressure while holding station, and takes the rotor noise up with it. So the controller climbs first and gives ground only when climbing has not been enough. This is the one thing a drone can do that a drover cannot, and it is worth building the whole standoff rule in slant range to get it.
• The mob is led by its slowest member. Anchoring on the middle of the mob walks the front half away and leaves the back half standing, which is how you arrive two short.

Read animals, not averages

A hundred head walking at a perfectly legal pace with one animal bolting inside them averages out to a compliant move. The one animal is the one that goes through a fence, and if it is a ewe it takes her lambs with it. By the time the average crosses a threshold the mob is already running.

So every animal is scored on how it is moving, from three signals that are only meaningful together:

Measure against the median of the mob, not the mean. The mean is dragged up by exactly the animals you are looking for, so a mob with several of them quietly raises its own bar and stops detecting any of them.

Hardware

The speaker earns its place more than anything else on that list. It is the cheapest item and it is what lets the aircraft work at 25 m instead of 10.

In-house software

The simulator is not a nice-to-have. A controller that cannot be replayed has no business near livestock, so nothing in the control path may use a clock or a random number: same seeds, same run, every time. It is published at the playground precisely so the claim can be checked rather than taken on trust, and one of the scenarios there is built for the controller to fail.

Cost efficiency

• Visual line of sight first, dock second, beyond line of sight third. The same order as the livestock patrol, for the same reason: each step multiplies the regulatory cost.
• Herding rides on the patrol's aircraft. Same airframe, same zoom and thermal, one extra payload. If you have built for livestock and fences, this is a speaker and software.
• The gates are a one-off survey. Route planning is cheap once somebody has walked the holding and recorded where the fences open and what the stock must be kept off. Until that is done the planner assumes the gate is midway between two blocks and says on the plan that it is assuming.
• Sell the record, not the labour saved. The twenty minutes is real, but what a farmer is buying is a move that is answerable: what it did, against what it was allowed to do, on every mob, every time.