An LSTM-based Bilateral Active Estimation Model for Robotic Teleoperation with Varying Time Delay
Xuhui Zhou, Weibang Bai, Yunxiao Ren, Ziqi Yang, Ziwei Wang, Benny Lo, Eric M. Yeatman
Abstract
The time delay of signal transmission in the bilateral teleoperation system remains to be a severe problem for stability and transparency, despite that various methods have been proposed for alleviating the effects. Among those approaches, the neural networks (NN) based method is model-free and adaptable for system uncertainties and disturbances, which have shown great potential for teleoperation signal prediction. In this study, firstly, the concept of Passive Prediction and Active Prediction is clarified, i.e., Passive Predictor makes prediction given the delayed signals, while Active Predictor gives prediction based on raw signals before transmission. Secondly, a new Long Short-Term Memory (LSTM) based Bilateral Active Estimation Model (BAEM) is proposed for estimating the time delay in both directions of the teleoperation system, and the condition and proof of the model stability are provided. Based on the proposed model, another LSTM-based Active predictor is used thereafter, to predict the teleoperation signals with as long time in advance as the estimated time delay provided in its transmission direction. The proposed prediction method is independent of dynamic systems, hence it is applicable for general teleoperation scenarios. Moreover, any predictive methods other than LSTM can be embedded in the model, showing great extensibility.