Robust Control

Control design that guarantees stability and performance despite model inaccuracy, bounded parameter variation, and external disturbance.

Robust control is a branch of control theory that took shape from the late 1970s onward. Its core idea is to admit model error up front, at design time: given a bounded range of uncertainty — say, payload mass within some interval, or unmodeled joint flexibility or friction — a single fixed controller is designed that guarantees stability and a performance floor across the entire range. Representative methods include H∞ control, pioneered by George Zames and others, which minimizes the worst-case effect of disturbances on the output, and sliding mode control. Robust control differs from adaptive control in that its controller parameters are fixed, relying on built-in margin to withstand uncertainty, whereas adaptive control estimates parameters online and adjusts the controller while running. The trade-off is that robust designs tend to be conservative, and still need a roughly accurate model to start from. In robotics, payload variation, joint friction, and varying ground stiffness are typical sources of uncertainty; domain randomization in reinforcement learning, which trains policies to be insensitive to parameter variation, pursues a similar kind of robustness.

ExampleAn arm must handle a workpiece weighing anywhere from 0 to 3 kg. Robust control designs a single controller up front assuming the load could be any value in that range, guaranteeing stability and bounded tracking error in every case. Adaptive control instead estimates the actual mass online after the object is grasped and adjusts its control parameters accordingly.

Also called

Robust Control Theory

Related

Adaptive Control · Sliding Mode Control · Lyapunov Stability · Disturbance Observer · Active Disturbance Rejection Control · Domain Randomization

Sources

Wikipedia: Robust control
Wikipedia: H-infinity methods in control theory