A drone adjusts its own weight and tilt mid-flight without manual recalibration.

Drones like quadrotors struggle to fly smoothly when carrying unknown or suddenly changing weights. The researchers created a smarter autopilot to solve this. Their system predicts how the drone will move by using mathematical approximations, known as Taylor expansionsTaylor expansionsA math method to approximate complex behavior using simpler polynomials., calculated up to the third order around a hover state. It learns the drone's behavior row by row as it flies, using updated data to instantly estimate the vehicle's mass and the exact upward thrust needed to counter gravity. This means the drone does not rely on a fixed, pre-set gravity setting. In simulations with sudden payload shifts and sharp spiral flight paths, the higher-order models improved flight predictions and path tracking while keeping the underlying control structure simple.

Why it matters

For drones carrying changing loads, like delivery robots, this means they can keep flying accurately without stopping. It reduces crash risks when weights shift suddenly.

Questions answered