Towards Probabilistic and Risk-Aware Active Disturbance Rejection Control: Particle-Filter-Based Extended-State Estimation and Control Action Selection

Abstract

This paper proposes a novel probabilistic approach to Active Disturbance Rejection Control (ADRC) based on particle filtering (PF) for extended-state estimation. Unlike the classical Extended State Observer (ESO), the proposed method represents the system state and total disturbance using weighted particles, enabling the direct incorporation of non-Gaussian noise and model uncertainty. The PF-based estimator is integrated within the ADRC framework and compared with conventional ESO- and Kalman filter-based approaches. In addition to the standard PF-ADRC scheme based on the posterior mean, a risk-aware control mechanism is introduced. The particle-wise state and disturbance realizations are used to construct a distribution of candidate control actions, which is subsequently employed to account for unfavorable posterior realizations and reduce excessive control activity and actuator saturation risk. Simulation studies conducted for nonlinear mechanical systems, including a preliminary validation on a higher-order nonlinear underactuated plant, illustrate the feasibility of the proposed framework under non-Gaussian noise, measurement outliers, nonlinearities, and actuator constraints. The results characterize the potential benefits, limitations and performance trade-offs of posterior-based action selection.