Beyond Wave Variables: A Data-Driven Ensemble Approach for Enhanced Teleoperation Transparency and Stability

TL;DR

The paper tackles time-delay challenges in bilateral teleoperation by replacing the traditional wave-variable transform with a data-driven stacking ensemble of three hybrid models optimized via Optuna and constrained to preserve passivity. The Prophet-LSTM, CNN-LSTM, and LSTM-KMeans-RF base learners feed an XGBoost meta-learner, forming a two-level architecture that learns to reconstruct ideal communication signals under variable delays and noise. Rigorous stability and passivity validations, including Lipschitz analysis and energy-based checks, demonstrate that the learned channel maintains safety while achieving transparency comparable to the baseline. Experimental results in MATLAB/Simulink show high predictive accuracy and strong stability guarantees, indicating that a learnable, stability-focused surrogate can be a viable alternative to wave-variable-based links in teleoperation. This work lays groundwork for hardware-in-the-loop validation, multi-DoF extensions, and online adaptation to dynamic network conditions.

Abstract

Time delays in communication channels present significant challenges for bilateral teleoperation systems, affecting both transparency and stability. Although traditional wave variable-based methods for a four-channel architecture ensure stability via passivity, they remain vulnerable to wave reflections and disturbances like variable delays and environmental noise. This article presents a data-driven hybrid framework that replaces the conventional wave-variable transform with an ensemble of three advanced sequence models, each optimized separately via the state-of-the-art Optuna optimizer, and combined through a stacking meta-learner. The base predictors include an LSTM augmented with Prophet for trend correction, an LSTM-based feature extractor paired with clustering and a random forest for improved regression, and a CNN-LSTM model for localized and long-term dynamics. Experimental validation was performed in Python using data generated from the baseline system implemented in MATLAB/Simulink. The results show that our optimized ensemble achieves a transparency comparable to the baseline wave-variable system under varying delays and noise, while ensuring stability through passivity constraints.

Figures (8)

• Figure 1: Overview of our problematic, research questions and the proposed methodology.
• Figure 2: The Improved four-channel teleoperation architecture, adapted from Chen et al.Chen2018
• Figure 3: Compensated Communication Channel with Modified Wave Transform.Chen2018
• Figure 4: The complete data processing and model training pipeline. The workflow begins with data generation in MATLAB/Simulink, followed by preprocessing and splitting. Base models are trained using K-fold cross-validation to generate meta-features, which are then used to train the final meta-learner.
• Figure 5: Visualization of the dataset splits across the full time-domain signals for all inputs and outputs. The main division into the Training Set (85%) and Test Set (15%, light blue) is shown. Within the training set, the 5-fold cross-validation splits are depicted by pastel-colored vertical bands. Finally, the internal 10% validation set used for training each neural network (e.g., for early stopping) is shown with a hatched red overlay.