Project Title: : Edge Intelligence for Model-Augmented Haptic Teleoperation in Tactile Internet
Description
Model-Augmented Haptic Teleoperation, also known as Model-mediated teleoperation (MMT) is a promising approach for Tactile Internet to address both stability and transparency in teleoperation under communication latency and packet loss. The idea of MMT is that the local device uses a local model, i.e., the digital twin of the remote device, to approximate the remote environment. Instead of transmitting actual haptic feedback from the remote device to the local device, the haptic feedback can be computed based on the local model without noticeable delay. However, accurate online model parameters estimation is challenging, model mismatch can, for example, cause position tracking error, resulting in dangerous slave behaviour. This project aims to design Edge Intelligence using transfer learning and continual learning approaches to develop a new neural network-based estimation of remote environment for MMT. The key is to design and implement new machine learning methods for achieving high performance with small training data, and to design and implement continual learning methods to improve the NNs’ performance based on newly collected experience. The outcome is expected to achieve model estimation acceleration for remote environment, and to increase model accuracy and minimize model mismatch, i.e., enhancing stability and transparency, under large latency.
