Watch the herding controller work
This is not an animation of how drone herding might go. It is the same controller that would fly the aircraft, run against a modelled paddock, and every position you see is a tick of it. Change the scenario and the controller has to solve a different problem. One of them it cannot solve, and it stops.
The two rings around the aircraft are the whole control story. The outer one is the distance it is keeping from the nearest animal; the inner one is how much of that distance is height rather than ground. When the mob unsettles, watch the inner ring grow while the aircraft stays where it is.
The road on the headland
A public road runs along the top of both blocks. The corridor has to be planned around it and the mob steered off it, because an animal on a road is the outcome this whole capability exists to prevent.
45 of 55 on the new block after 5 min 32 s.
9 left on the old block, reported for somebody to walk out to.
What it did, against what it was allowed to do
- Closest the aircraft came limit 25 m standoff
- 24 m
- Fastest the mob moved limit 1.6 m/s
- 0.85 m/s
- Highest the aircraft worked limit 110 m ceiling
- 28 m
- Most of the mob agitated at once limit 8 per cent
- 11 per cent
- Worst individual animal
- 0.75
- Times it eased off
- 2
- Time under pressure limit 20 min
- 5 min 32 s
- Route clearance from the road
- 54 m
Briefly closer than the standoff, because an animal walked toward it and the aircraft took a moment to yield. Inside the tolerance, and the move would have stopped had it persisted.
Over the line for a stretch, which is what sent the aircraft up. The limit stops a move only if it is still over it a minute later, because climbing is the remedy and climbing takes time.
A score out of one, from how that animal was moving. This is the reading the mob average cannot see.
Height comes off first
A stockman under pressure can only step back, and a mob that is backed away from stops walking. An aircraft can step up instead: it sheds the pressure, takes the rotor noise with it, and keeps the position that is doing the driving. So the controller climbs before it gives ground, and only gives ground when climbing has not been enough.
It watches animals, not averages
A hundred head walking at a legal pace with one animal bolting inside them averages out to a perfectly compliant move, and the one animal is the one that goes through a fence. Every animal is scored on how it is moving: fast relative to the mob, turning hard, or leaving. That reading is what sends the aircraft up.
Stopping is a result
One of these scenarios cannot be solved inside the limits a mob may be driven under, and the controller stops and asks for a person. That rung is on this page on purpose. A demonstration that only shows the good case is an advertisement, and an advertisement is not evidence.
Why you can check this
Every scenario is generated deterministically: a fixed seed, no clock, no random numbers. The
same run produces the same frames every time, which is what lets the figures above be asserted
in the test suite rather than quoted from a good day. Run it yourself with
buddy herding:sim --scenario road-hazard.
The animals are a model, not a claim about livestock. What is real here is the controller: the same module decides where the aircraft goes on this page, in the tests and on a flight. See the capability for how a move is planned, authorised and flown.