A driver on a slippery road doesn't know exactly how the car will respond to the steering wheel, and learns on the go, leaving room for error. A smart control algorithm behaves in a similar way. Instead of requiring a precise model, it collects data on the fly, estimates uncertainty and external disturbances, as if a black box gradually reveals its settings.
Based on these estimates, the algorithm charts a safe corridor — something like a virtual lane that narrows or widens depending on its confidence. Commands are given with a margin, so even in the worst case the system does not exceed the boundaries.
The main advantage is guarantees from the very first step. There’s no need to wait for the robot to accumulate statistics: the operation remains stable right away. Numerical tests confirmed this: a virtual drone maintained its balance despite sudden jolts. Such adaptability paves the way for drones and fast robots that previously required lengthy tuning.
Model predictive control (MPC) was originally used in chemical plants, where processes are slow and everything can be calculated in advance. Now it learns on the fly, opening the way for fast robots and drones.