Stable Neural Adaptive Filters for Teleoperations With Uncertain Delays

Uncertainties in communication networks negatively affect the performance and usability of teleoperation systems, specially, in time-critical applications such as telesurgery. There already exist different methods to tackle this problem using filtering and learning approaches to smoothly estimate perturbed reference signals. Despite these efforts, the instability issue remains unsolved for such systems under random time-delay perturbations. This study employs and extends one of the best filtering techniques and proposes a new strategy based on learning capabilities of artificial neural networks to adaptively filter out delay-related disturbances and provides the most stable yet accurate estimations. To achieve this goal, an adaptive learning mechanism is proposed based on a Lyapunov-Krasovskii functional to not only analyse and guarantee the stability of the overall system, but also preventing the learning algorithm to get stuck in local optima. The proposed method is experimentally evaluated and compared with other closely similar methods in the recent literature. The results demonstrate the outstanding performance of the proposed solution in this study.