Practical robotic systems require control methods that remain reliable under limited computational resources, uncertain environments, and frequent changes in operating conditions. Although model-based control forms the foundation of high-performance robotics, real-world deployment is often hindered by model uncertainty, time-varying dynamics, and costly identification. As a result, low-order and intuitive control schemes remain dominant, yet such approaches often fail to sustain consistent performance under disturbances and parameter variations. Robust and adaptive control provide representative paradigms to address this gap, where a Disturbance Observer (DOB) suppresses uncertainty through disturbance rejection and a Parameter Adaptation Algorithm (PAA) improves model fidelity through online identification. However, direct integration of a DOB and a PAA often introduces functional interference, including mutual masking between disturbance compensation and parameter estimation, which compromises closed-loop stability. This paper proposes an Adaptive Disturbance Observer (ADOB) that integrates a DOB with online parameter adaptation. The ADOB updates the nominal model of the DOB in real time using a Recursive Least Squares (RLS)-based PAA, while a dual-filtering structure separates disturbance rejection and parameter identification. Stability is analyzed using hyperstability theory, where a smoothing mechanism enforces the slowly varying parameter assumption. Experiments on a one-Degree-of-Freedom (DOF) electromagnetic actuator and a three-DOF robotic manipulator demonstrate reductions in model uncertainty and tracking error compared with a conventional DOB.
Park et al. (Wed,) studied this question.