Machine Learning-Based Performance Improvement of Bilateral Teleoperation with Hydraulic Actuator

Teleoperation with hydraulic actuator is useful for human action augmentation. However, disturbances in hydraulic actuators are complex and accurate estimation of external forces is difficult. In this paper, a reaction force observer and machine learning are combined to achieve high accuracy sensorless force estimation in hydraulic actuator. Furthermore, this method is applied to a bilateral control system to improve its performance. While there are many machine learning methods, this paper uses a Long Short-Term Memory network, a type of recurrent neural network that excels at inferring time series data, to accurately infer the hysteresis characteristics of disturbances in hydraulic actuator. Furthermore, 4ch bilateral control based on oblique coordinate control is used to realize teleoperation. In the experiment, a friction model-based compensation method and a machine learning-based compensation method are applied to bilateral control, and the performance of each method is evaluated.