Why GPS is not enough
GPS needs a clear view of the sky. Indoors, in warehouses, tunnels and between tall buildings, the signal is weak or blocked, so a drone that relies on GPS alone can drift or lose its position completely.
Three sources of position
An inertial measurement unit (an accelerometer and a gyroscope) senses movement hundreds of times a second, but its error grows over time. A downward camera can track the texture of the floor to estimate speed, which is called optical flow. Range sensors such as lidar or time-of-flight sensors measure height and the distance to walls and obstacles.
Seeing the world
Visual odometry follows features from one camera frame to the next to work out how the drone has moved. SLAM (simultaneous localisation and mapping) goes further: it builds a map while working out where the drone is inside it. A simpler approach uses printed markers, known as fiducial markers, such as ArUco tags. Because the marker's size and shape are known, a single camera image tells the drone its position relative to the marker.
Putting it together
Real systems combine all of these with sensor fusion, usually a Kalman filter, so that each source covers the weaknesses of the others. Lighting, motion blur and limited onboard computing power are the hard parts, so most of the work is making the vision robust and fast enough to run on the drone itself.
Where Aerosoul fits
Aerosoul is prototyping vision-guided flight through tight courses without satellite help. This is early-stage work, funded by our apps, and one part of our wider robotics plans.
