Guide
How a drone finds its way indoors.
Outdoors, a drone asks the satellites where it is. Inside a home there are no satellites to ask, so it has to work it out from its own senses, many times a second, while the furniture, the people and the air keep changing around it.
In short. An indoor drone combines fast motion sensors with slower senses that look at the room: a downward camera that measures speed against the floor, and distance sensors that measure how far the floor, walls and furniture are. A filter blends them into one estimate, and a map of what it has seen lets it correct the drift that builds up. Then a planner finds a path through free space, under rules: do not fly towards what you have not seen, keep a margin from what you cannot see well, and stop when unsure.
Three questions, all the time
An indoor robot keeps answering three questions: where am I, what is around me, and where may I go next. Each has its own senses, and none is perfect on its own.
Where am I?
- Motion sensors. A gyroscope measures turning and an accelerometer measures acceleration, hundreds of times a second and without looking at anything, but their small errors add up quickly.
- A camera looking at the floor sees how fast the floor slides past, like an optical mouse, and turns that into speed once it knows its height. It needs texture and light: optical flow.
- Distance sensors time light to measure how far the floor, the walls and the furniture are; small multi-zone sensors measure a grid of distances at once: time-of-flight sensors.
A filter weighs these against each other many times a second, trusting each where it is strong: what a Kalman filter does. Combined, they give a good estimate for a while. Then the small errors grow: this is drift. A robot that flies for a few minutes may be off by several centimetres unless it corrects itself.
What is around me?
As it flies, the robot builds a map of what its sensors have seen: free space, walls, furniture, and things it should treat with care, such as glass. Comparing what it sees now with its map is also how it corrects its drift: what SLAM is. Glass and mirrors are the classic trap, because light passes through or bounces off them; sound reflects from glass, so a second sense helps: glass and mirrors. A camera that recognises people, pets and curtains adds what distance alone cannot tell: how object detection works.
Where may I go?
A route planner finds a path through the free space on the map. Good rules matter as much as good maps: do not fly towards space you have not seen, keep a margin from what you cannot see well, and stop when you are unsure. The margin should grow with the robot’s own uncertainty: a robot that is less sure where it is should stay further from walls and people.
When the room fights back
- Moving air. A fan or an air conditioner pushes a light drone off its path, and the controller has to notice and push back: what moving air does.
- Narrow gaps. A doorway left half open is a test of everything above at once; a drone that can fold its arms gets through narrower gaps: foldable drones and through a narrow door.
- Landing. Coming down needs the most care, because the floor, a table or a person’s things are close: how a robot chooses where to land.
Tested in simulation first
Much of this is tried in simulation before anything flies, because a simulated room can throw thousands of situations at a robot in an afternoon. A simulation is not the world, though, and the difference has a name: the reality gap.
In Alkyon Concept
Alkyon is designed to see only through its own sensors, build its map of your home as it flies, and refuse to fly towards space it has not seen. In simulation, the hard part is exactly the drift above; correcting it against the map is one of our open tasks. What else goes wrong indoors: flying a drone indoors. How the propellers keep it up in the first place: how quadcopters fly.
Questions
Why not use GPS indoors?
Satellite signals are weak inside buildings and bounce off walls; even outdoors they are accurate to metres, and a robot passing a doorway needs centimetres.
What sensors does an indoor drone need?
At least motion sensors (a gyroscope and an accelerometer) and something that measures its speed or position against the room, usually a downward camera for optical flow and distance sensors for the floor and the walls. Sensors that use light miss glass, so some robots add sound.
What is drift?
The slow growth of a robot’s position error as small sensor errors add up. Correcting it means recognising something already on the map, such as a wall or a doorway, and pulling the estimate back.