Model-Reference Adaptive Control

MRAC adjusts controller parameters online so the closed loop matches a chosen reference model despite plant uncertainty.

Making the Plant Behave Like a Model

Model-reference adaptive control (MRAC) specifies desired performance through a reference model, a stable system whose response is what the closed loop should imitate. An adaptation mechanism then tunes the controller's parameters in real time so the plant's output converges to the reference model's output, even when the plant is poorly known.

The architecture

• Reference model: encodes the target dynamics, driven by the same command as the plant.
• Adjustable controller: a feedback and feedforward structure with parameters that adaptation will set.
• Adaptation law: updates the parameters based on the error between plant output and reference-model output.
• The tracking error drives adaptation until the plant matches the model.

The MIT rule

The original adaptation law, the MIT rule, adjusts each parameter in proportion to the negative gradient of the squared output error with respect to that parameter. It is intuitive and easy to implement, but it carries no general stability guarantee: with large adaptation gains or fast commands it can itself become unstable.

Lyapunov-based adaptation

The preferred modern approach derives the adaptation law from a Lyapunov function chosen so that its derivative is negative along the combined dynamics of tracking error and parameter error. This construction guarantees that the tracking error converges and the system stays stable, turning MRAC from a heuristic into a method with a proof.

Practical cautions

MRAC assumes the plant can, in principle, be made to match the reference model, which requires matching conditions on structure and relative degree. It is sensitive to unmodeled dynamics and disturbances, which can cause slow parameter drift toward instability. Robustifying modifications, such as bounding the parameters or adding leakage, are used to keep long-running adaptive loops safe.

MRAC has a long aerospace heritage and remains a template for adaptive design, illustrating clearly how a stability-derived learning law lets a controller cope with a plant it does not fully know.