Abstract
Bilateral teleoperation systems are vital for performing complex tasks in hazardous or remote environments. However, ensuring stability and transparency—where the operator accurately feels remote forces—remains a challenge due to communication delays and system uncertainties. This paper presents a robust control strategy combining Radial Basis Function Neural Networks (RBFNN) with Adaptive Sliding Mode Control (ASMC) and a novel projection mapping technique to maintain performance under time-varying delays.
The Challenge of Bilateral Teleoperation
In a bilateral teleoperation system, a human operator interacts with a "master" device to control a "slave" robot in a remote environment. The system's goal is transparency: the operator should perceive the remote environment as if they were physically present.
Two primary obstacles hinder this goal:
- Transmission Delays: Lag in the communication channel can lead to oscillations or total system instability.
- Dynamic Uncertainties: Nonlinearities like friction and gravity in the robotic hardware are difficult to model precisely, leading to tracking errors.
Proposed Control Methodology
The researchers developed a multi-layered architecture to address these issues simultaneously. The core of the approach is the integration of RBFNNs to approximate unknown system dynamics, coupled with an Adaptive Sliding Mode Controller for robust tracking.
Key Components:
- RBFNN Approximator: Acting as a universal function approximator, the RBFNN learns and compensates for modeling errors and external disturbances in real-time.
- Adaptive Sliding Mode Control (ASMC): This controller provides the necessary "muscle" for high-speed, robust tracking, with adaptive gains that respond to the RBFNN's estimates.
- Reference Trajectory Producers (RTP): Implemented on both sides, RTPs generate passive reference paths, ensuring the system remains energetically stable (passive) despite delays.
- Projection Mapping: A critical innovation is the use of a saturation-based projection mapping on the master side. This bounds the delayed environmental torque signal, preventing it from destabilizing the master controller during high-lag periods.
Experimental Results
The system was validated using two Phantom Premium haptic devices. Experiments simulated a challenging environment with a time-varying delay of 0.4s ± 0.1s and artificial disturbances.
The proposed controller was compared against a standard PD controller. While the PD controller became unstable within 7 seconds under these conditions, the proposed RBFNN-ASMC system maintained perfect stability. Position tracking errors were kept exceptionally low (approx. $10^{-5}$ rad), and force feedback remained highly accurate ($10^{-4}$ Nm), demonstrating superior transparency.
Conclusion
By combining neural network approximation with robust adaptive control and passivity-based trajectory generation, this research provides a viable path for stable, high-fidelity remote operation. The inclusion of projection mapping effectively decouples the threat of instability from communication latency, a significant step forward for teleoperation in unpredictable network environments.
Summary updated