HMFP-DBRNN: Real-Time Hand Motion Filtering and Prediction via Deep Bidirectional RNN

Soroosh Shahtalebi, Seyed Farokh Atashzar, Rajni V. Patel, Arash Mohammadi

Research output: Contribution to journalArticle

Abstract

Pathological hand tremor (PHT) is among the most common movement symptoms of several neurological disorders including Parkinson's disease and essential tremor. Extracting PHT is of paramount importance in several engineering and clinical applications such as assistive and robotic rehabilitation technologies. In such systems, PHT is modeled as the input noise to the system and thus there is a surge of interest in estimation an compensation of the noise. Although various works in the literature have attempted to estimate and extract the PHT, in this letter, first, we argue that the ground truth signal used in existing works to optimize the performance of tremor extraction techniques is not accurate enough, and thus the performance measures for the prior techniques are not perfectly reliable. In addition, most of the existing tremor extraction techniques impose unrealistic assumptions, which are, typically, violated in practical settings. This letter proposes a novel technique that for the first time incorporates deep bidirectional recurrent neural networks as a processing tool for PHT extraction. Moreover, we devise an intuitively pleasing training strategy that enables the network to perform not only online estimation but also online prediction of the voluntary hand motion in a myopic fashion, which is currently a significantly important unmet need for rehabilitative and assistive robotic technologies designed for patients with pathological tremor.

Original languageEnglish (US)
Article number8618351
Pages (from-to)1061-1068
Number of pages8
JournalIEEE Robotics and Automation Letters
Volume4
Issue number2
DOIs
StatePublished - Apr 2019

Keywords

  • Deep learning in robotics and automation
  • medical robots and systems
  • rehabilitation robotics

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Biomedical Engineering
  • Human-Computer Interaction
  • Mechanical Engineering
  • Computer Vision and Pattern Recognition
  • Computer Science Applications
  • Control and Optimization
  • Artificial Intelligence

Fingerprint Dive into the research topics of 'HMFP-DBRNN: Real-Time Hand Motion Filtering and Prediction via Deep Bidirectional RNN'. Together they form a unique fingerprint.

Cite this