Active Disturbance Rejection Control
- Introduced by
- Jingquan Han
- Year
- 2009
- Type
- algorithm
- Family
- Robust Control
What does it do?
Active Disturbance Rejection Control (ADRC) is a control method that estimates and cancels disturbances and model uncertainties in real-time using an extended state observer (ESO), treating them as additional 'disturbance states'.
Developed by Han and popularized by Gao, ADRC achieves remarkable robustness without requiring precise plant models, making it practical for real-world systems with significant uncertainty and disturbances.
The intuition behind the method
ADRC views any unmodeled dynamics, disturbances, and parametric errors as an additive 'lumped disturbance'. Rather than tuning a robust controller for worst-case disturbance, ADRC estimates this disturbance in real-time using an extended state observer (ESO) and then cancels it through feedback. The ESO augments the state space with a disturbance state that is estimated alongside traditional states. This estimated disturbance is subtracted from the control law, leaving a linearized, disturbance-decoupled system that can be controlled easily. The approach is pragmatic: embrace model uncertainty, estimate it, and compensate.
History
Jingquan Han introduced ADRC in the early 2000s in China; it gained recognition internationally around 2006. Gao's work popularized ADRC in the U.S. control community, highlighting superior performance on benchmark problems. Industrial adoption grew in 2010s, especially in China and Asia. Modern variants include adaptive ADRC, distributed ADRC, and learning-augmented ADRC.
When to use it
Use ADRC when the plant model is inaccurate, disturbances are substantial, and you want simple control law without detailed model. Ideal for industrial applications (motors, pumps, heating), robotics with uncertain payloads, and systems with time-varying parameters. Avoid ADRC if the disturbance is too fast for the observer to track or if very low steady-state ripple is required.
Limitations and considerations
- Requires accurate output measurement; observer sensitivity to noise can limit performance.
- Observer bandwidth limited by measurement noise; cannot estimate arbitrarily fast disturbances.
- High observer gain can amplify measurement noise; trade-off between disturbance estimation and noise rejection.
- Tuning ESO parameters (observer gains) requires experience; less systematic than classical control methods.
- Theoretical analysis more complex; stability and convergence guarantees less mature than classical methods.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Robust to model uncertainty; treats uncertainty as estimated disturbance rather than designing for worst-case.
- Simple control law (often PD-like); minimal tuning compared to PID or H-infinity.
- Fast disturbance rejection; real-time estimation enables rapid response.
- Extends to MIMO and nonlinear systems naturally.
- Practical implementation; no need for detailed model identification.
Applications
- Industrial drives: Motor speed control with load variations and friction uncertainty.
- Positioning systems: Precise positioning of stages despite friction and mechanical play.
- HVAC: Temperature control with disturbances from outdoor conditions and occupancy changes.
- Renewable energy: Wind turbine control adapting to varying wind and turbulence.
- Robotics: Arm control with uncertain loads, friction, and cable effects.
How it works
How it works
Run an extended state observer (ESO) that estimates state x and lumped disturbance d from measurements; ESO is a bank of high-gain observers.
The estimated disturbance d_hat is fed into the control law; subtract it: u = u_nominal − d_hat, effectively canceling disturbance impact.
Design a simple linear controller for the disturbance-compensated system; often a proportional-derivative (PD) law suffices.
Close the loop: apply u to system, measure output, update ESO and control law; repeat.
Common pitfalls
- Setting observer gains too high; leads to noise amplification and instability.
- Using ADRC for systems with fast unmodeled dynamics; ESO bandwidth may be insufficient.
- Ignoring measurement quality; poor sensor signal degrades observer and overall performance.
- Assuming all disturbances are observable; if disturbance does not directly affect measured output, ESO cannot estimate it.
Frequently asked questions
How do I tune the extended state observer (ESO) gains?
ESO gains control observer bandwidth and robustness to noise. Higher gains estimate disturbance faster but amplify measurement noise. Common tuning: start with observer bandwidth 5-10x controller bandwidth, adjust empirically based on sensitivity to measurement noise. Parameter space is large; use frequency response analysis or pole placement for systematic design.
Related methods
Sources
Han, J. (2009). From PID to active disturbance rejection control. IEEE Transactions on Industrial Electronics, 56(3), 900-906.
Open DOI referenceCopied.Gao, Z. (2006). Active disturbance rejection control: a paradigm shift in feedback control system design. Proceedings of the 2006 American Control Conference, 2652-2661.
Open DOI referenceCopied.