| Summary: | The usefulness of medical devices, which require user input, is often limited by the control schemes that operate them. The recognition of user intent could enable far more intuitive control schemes that respond automatically to what the user wants the device to do. This paper provides a definition for intent, and then aims to systematically review current methods for sensing intent. It compares the accuracy of different methods and discusses how they might be combined. A systematic literature search was performed using IEEE Xplore, PubMed and Web of Science databases. 2311 papers were considered, reduced to 155 after review. All selected papers were assessed for quality using a checklist. The results identified and compared 15 sensing modalities used for intent sensing in a range of situations and applications that broadly fell into 12 distinct categories, with highly varying levels of accuracy. Several papers reached accuracy levels that could be suitable for everyday clinical application, but most work done on intent sensing to date has focused on activity transition classification, with fewer papers addressing task goal interference or predicting future actions. Further work can focus on the implementation of these kind of methods into a combined, context-aware intent-sensing control system.
|
|---|