IPT-DCD: Interpolation Predictor for Teleoperation Under Dynamic Communication Delay Using Deep Learning Approach

Teleoperation systems experience degraded control stability and safety due to dynamic communication delays. This study proposes an Interpolation Predictor for Teleoperation under Dynamic Communication Delay (IPT-DCD), a predictor that reconstructs asynchronously received control commands via interpolation and predicts future commands using an encoder-decoder LSTM architecture. To restore the temporal consistency of delayed signals, a signal preprocessing technique called the Backward Shifting and Interpolation (BSI) was applied, enabling the transformation of received data into an undelayed and uniformly sampled format. As a result, the proposed model was capable of generating real-time steering command outputs through a many-to-many time series structure. Furthermore, to evaluate its effectiveness, IPT-DCD was experimentally compared with a baseline model, a Predictor for Teleoperation under Dynamic Communication Delay (PT-DCD). The results reveal that IPT-DCD exhibits significantly greater robustness to large communication delay outliers than the baseline, highlighting its effectiveness in dynamic and unstable teleoperation environments.