💻 computer science

Output Feedback Adaptive Performance Control

This paper proposes a robust output-feedback adaptive controller for uncertain high-order nonlinear systems with actuator constraints and unavailable state measurements, utilizing a novel Prescribed Performance Observer and an adaptive performance adjustment mechanism to guarantee bounded signals and recover state-feedback tracking performance.

Original authors: Panagiotis S. Trakas, Charalampos P. Bechlioulis

Published 2026-08-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Panagiotis S. Trakas, Charalampos P. Bechlioulis

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the world of robotics and autonomous vehicles, machines are constantly trying to follow a path, whether it is a drone tracking a moving target or a robot arm assembling a delicate component. The challenge lies in the fact that these machines are often uncertain about their own internal state; they cannot always see their own speed or acceleration, only their current position. Furthermore, the physical motors that move them have hard limits. They cannot spin infinitely fast, nor can they push with infinite force. If a controller asks a motor to do the impossible, the machine stumbles, loses its way, or even breaks. For decades, engineers have struggled to design a "brain" for these machines that can keep them on track despite not seeing the whole picture and while respecting the strict physical limits of their muscles. The goal is to guarantee that the machine stays within a safe, shrinking zone of error, getting closer and closer to the target without ever crashing into the walls of its own limitations.

A team of researchers has now developed a new control strategy that solves this difficult puzzle for a wide class of complex machines. Their work focuses on systems where the desired path is not a pre-written script but is measured in real-time, like a car chasing another car that is changing direction. In such situations, the machine cannot rely on knowing the future path or its own hidden internal speeds. The researchers created a robust system that uses only the visible output—the current position error—to steer the machine. They introduced a new type of observer, a mathematical tool that acts like a set of eyes, estimating the hidden speeds and accelerations with high precision. Unlike older methods that often amplify noise or require massive, unstable gains to work, this new observer adjusts its own sensitivity dynamically. It becomes sharper when the error is large and calms down as the machine settles, ensuring that the estimates remain accurate without being thrown off by the static and interference common in real-world sensors.

The core of this innovation is a mechanism that adapts the machine's performance goals to its available power. Imagine a runner who is told to reach a finish line in a specific time. If the runner is healthy and strong, they sprint. But if they are injured or carrying a heavy load, the goal adjusts; they are still expected to finish, but the timeline stretches to match their current capability. Similarly, this new controller constantly monitors how close the machine is to hitting its motor limits. If the motors are struggling, the controller temporarily relaxes the speed at which the error must shrink, preventing the system from demanding more power than the motors can give. This prevents the machine from stalling or becoming unstable. Once the motors have room to breathe, the controller tightens the requirements again, driving the error down to a tiny, pre-defined size. This dynamic adjustment ensures that the machine always operates within its physical safety zone while still achieving the best possible performance.

To prove their idea works, the researchers tested it on two very different physical models: a flexible robot arm with a joint that bends like a spring, and the rolling motion of a delta-wing aircraft. In the first test, the robot arm had to track a sine wave path while its motors were limited in both how hard they could push and how fast they could change that push. The new controller kept the arm's tracking error well within the safe boundaries, even as the motors hit their limits. The estimated speeds and positions matched the reality closely, and the control signals applied to the motors were smooth, avoiding the jerky, erratic movements that often plague other systems. In the second test, involving the aircraft, the system had to handle a rolling motion that naturally tends to grow unstable. Even when the researchers reduced the maximum force the motors could apply to a very low level, the new controller kept the aircraft stable and on course. In contrast, older methods that did not adapt to these tight limits failed, causing the aircraft to spin out of control.

The study also examined how the system handles the static and noise that inevitably come from real-world sensors. When the researchers added random measurement noise to the data, the new controller continued to track the path accurately. More importantly, the control signals it sent to the motors remained remarkably smooth. Other methods, when faced with the same noise, produced jagged, vibrating commands that could wear out motors or damage delicate mechanisms. The researchers found that their approach achieved this smoothness without needing to know the exact mathematical details of the machine's physics or the nature of the disturbances. They demonstrated that by combining a smart, adaptive observer with a controller that respects the machine's physical limits, it is possible to achieve reliable, high-precision tracking even in the most uncertain and constrained environments. The work suggests that this framework could be a practical solution for the next generation of autonomous systems that must operate safely and efficiently in the real world.

Technical Summary: Output Feedback Adaptive Performance Control for Uncertain High-Order Nonlinear Systems

Problem Statement
This paper addresses the control of uncertain high-order nonlinear systems performing dynamic tracking tasks under two critical constraints: hard actuator limitations (amplitude and rate) and limited feedback availability. In the considered framework, only the system output error is measurable; the internal system states and the derivatives of the desired trajectory are unavailable for feedback. This scenario is typical in applications such as autonomous vehicles and robotic systems where the reference trajectory (e.g., a moving target) is determined in real-time via sensors, rendering its higher-order derivatives unknown. The challenge lies in designing a controller that guarantees user-defined transient and steady-state performance specifications while ensuring all closed-loop signals remain bounded, despite model uncertainties, measurement noise, and the inability to access full state information or trajectory derivatives.

Methodology
The proposed solution integrates a novel Prescribed Performance Observer (PPO) with an Adaptive Performance Control (APC) framework.

Adaptive Performance Control (APC): Building upon previous work, the APC methodology dynamically adjusts performance specifications to accommodate hard input constraints. Unlike static performance funnels, the proposed scheme introduces an input-dependent convergence rate function. This mechanism allows the system to leverage available control effort during transient phases to accelerate convergence toward the steady-state set when actuators are not saturated, while relaxing performance constraints when saturation occurs to ensure feasibility. The control law explicitly handles both amplitude ( $u \in [-\bar{u}, \bar{u}]$ ) and rate ( $\dot{u} \in [-\bar{r}, \bar{r}]$ ) constraints through a dynamic control law involving saturation functions and adaptive performance envelopes ( $\rho_u, \rho_r$ ).
Prescribed Performance Observer (PPO): To address the lack of state measurements, the authors design a nonlinear observer that extends the standard High-Gain Observer (HGO). The PPO employs time- and error-dependent dynamic gains. Specifically, it utilizes a performance function $r(t)$ and a transformation $T(\cdot)$ to inject correction terms based on the output estimation error.
- Structure: The PPO retains the computational simplicity of a standard HGO (order $O(n)$ ) but avoids the need for fixed, high gains that typically amplify measurement noise or cause peaking phenomena.
- Gain Dynamics: The injection gains vary dynamically. While the time-dependent coefficients decrease exponentially, the transformed injection results in a steady-state differential gain of order $O(n)$ , similar to a classical HGO, but with a smoother transient response.
- Complexity: The PPO requires one fewer state than a standard Extended State Observer (ESO) and has comparable computational complexity to an HGO.
Separation Principle and Stability: The paper establishes a separation principle, proving that the output-feedback scheme recovers the performance of its state-feedback counterpart. By selecting the observer's ultimate accuracy bound ( $r_\infty$ ) sufficiently small, the observer dynamics converge faster than the controller dynamics. Theoretical analysis, utilizing singular perturbation techniques and Lyapunov stability theory, demonstrates that all closed-loop signals remain uniformly ultimately bounded and that the tracking error converges exponentially to a predefined neighborhood of the origin, provided the actuator limits satisfy specific feasibility conditions.

Key Contributions

Novel PPO Design: Introduction of a nonlinear observer with time and error-dependent injection gains that provides fast transient error correction without requiring plant-model information, maintaining the low computational cost of standard HGOs.
Input-Dependent Convergence Rate: A departure from existing APC literature, this work proposes a convergence rate function that dynamically adjusts based on the available control effort. This enables the system to explicitly leverage control authority during transients to accelerate convergence under hard input constraints.
Unified Output-Feedback Framework: The development of a control architecture that integrates the PPO with APC under simultaneous hard amplitude and rate constraints. Unlike prior funnel-based output-feedback methods that rely on heuristic gain tuning or virtual saturation limits, this scheme provides rigorous feasibility conditions guaranteeing bounded signals and prescribed performance.

Simulation Results
The efficacy of the proposed scheme is validated through three simulation studies:

Flexible Joint Robot: A 4th-order system demonstrates accurate tracking and smooth control input application under amplitude and rate limits, with the PPO achieving high estimation accuracy after a short transient.
Wing Rock Motion Control: Applied to a delta-wing aircraft model, the controller successfully recovers state-feedback performance even under heavy input saturation ( $\bar{u}=0.7$ ), where a comparative HGO-based controller from existing literature fails due to internal instability. The results highlight the robustness of the proposed method against varying initial conditions and observer parameters.
Ship Course Control (Comparative Study): The proposed controller is compared against funnel control and other adaptive methods on a ship model. The proposed scheme achieves faster convergence and smaller steady-state errors without requiring prior knowledge of the input gain or trajectory derivatives.
- Noise Robustness: Under measurement noise, the proposed controller produces significantly smoother control signals compared to competitors, which exhibit chattering due to noise amplification. While the PPO's noise sensitivity is noted as a limitation requiring future work, the rate-limiting mechanism effectively mitigates adverse effects.

Significance and Claims
The paper claims to provide a unified solution for uncertain nonlinear systems that combines robustness, high-performance control, and rigorous handling of actuator constraints. The significance lies in the ability to guarantee adaptive performance specifications in the presence of feedback limitations and hard input constraints without relying on explicit system dynamics or disturbance models. The authors emphasize that the proposed framework ensures the "best feasible performance" within physical actuator limits by dynamically adjusting performance funnels. While the method introduces some conservatism by not leveraging explicit model knowledge, it offers a practical, robust alternative for real-world applications where model uncertainty and actuator saturation are prevalent. The work concludes that while the PPO effectively balances estimation accuracy and computational simplicity, future research is needed to further reduce its sensitivity to measurement noise.

More like this