Robot mowers without a wire, explained

RTK GPS, lidar, cameras and wheel counting, and why "no boundary wire" does not mean "no problems". Switch the sensors off and watch the mower lose itself.

For twenty years the answer to “how does the robot know where the lawn ends” was a wire. You buried it round the edge, the mower sensed it, and inside that loop it bounced about at random. It never knew where it was. It only knew when it had left.

The new mowers throw the wire away and promise stripes instead. To cut stripes you have to know where you are to within a few centimetres, all the time, under trees and beside walls. Here is how they try, and how they fail.

The four ways a mower can know where it is

RTK GPS. Ordinary GPS is out by a few metres, which is useless on a lawn. RTK adds a base station on your shed that watches the same satellites and works out the error, then tells the mower. With a clear sky the fix is about two centimetres. The catch is in the words “clear sky”: stand under a tree canopy, or close to a wall that bounces the signal, and the fix goes vague or vanishes. In the simulator, leave RTK as the only sensor and watch the lime ring swell the moment the mower goes under the tree.

Lidar. A spinning laser that measures distance to everything around it. It is superb next to a wall, a hedge or a tree trunk, because it can line those up with its map. On an open lawn there is nothing to see, so lidar alone drifts in the middle of the grass. Turn RTK off and lidar on and you will see the fix come and go as the mower passes the wall and the tree.

Cameras. The cheapest sensor and the cleverest software. The mower learns what grass looks like, where the flower bed starts, which blob is the cat. In daylight it holds a fix of a few centimetres almost anywhere. Then the light goes. Press Dusk.

Wheels and an IMU. Every mower has these. Count wheel turns, measure turning with a gyroscope, and you can dead-reckon your position with no outside help at all. The problem is that every turn adds a little error and it never gets taken away. In the simulator that is the slow swelling of the ring whenever nothing else has a fix.

Fusion is the actual product

No single sensor works everywhere, so the good mowers run all of them and trust whichever is best right now. RTK on the open lawn. Lidar along the wall. Camera under the tree. Wheels for the two seconds in between. The expensive part of a no-wire mower is not the motor or the blades. It is the software deciding, forty times a second, which sensor to believe.

Try switching everything on in the simulator. The ring stays small the whole way round. Now switch off one at a time and find the spot in your own garden where that mower would get lost.

What this means when you’re buying

• Walk your lawn and look up. Big trees, a tall house wall, a north-facing strip under a hedge. Each one is a place RTK alone will struggle. If that is half your garden, you want a camera or lidar as well.
• “RTK” on the box is the start of the question, not the end. Ask what it falls back on when the sky goes.
• Camera-only mowers mow in the day. That is fine, unless you wanted it to work at night so the garden is yours in the daytime.
• Bumps are not failures. Watch the edge-hit counter. A mower that touches the fence and corrects is doing the right thing. One that never bumps is either perfect or never leaves the open lawn.

Why a hedge is harder

Everything above gets worse when the thing being cut is vertical. A hedge trimmer that flies has to hold position to a blade’s width while hovering next to a wall of leaves that blocks half the sky and looks the same from every angle. That is the problem Kestrel is built around, and the reason this blog exists.