Gaze Estimation Technique for Directing Assistive Robotics

Abstract

Assistive robotics may extend capabilities for individuals with reduced mobility or dexterity. However, effective use of robotic agents typically requires the user to issue control commands in the form of speech, gesture, or text. Thus, for unskilled or impaired users, the need for a paradigm of intuitive Human-Robot Interaction (HRI) is prevalent. It can be inferred that the most productive interactions are those in which the assistive agent is able to ascertain the intention of the user. Also, to perform a task, the agent must know the user's area of attention in three-dimensional space. Eye gaze tracking can be used as a method to determine a specific Volume of Interest (VOI). However, gaze tracking has heretofore been under-utilized as a means of interaction and control in 3D space. This research aims to determine a practical volume of interest in which an individual's eyes are focused by combining past methods in order to achieve greater effectiveness. The proposed method makes use of eye vergence as a useful depth discriminant to generate a tool for improved robot path planning. This research investigates the accuracy of the Vector Intersection (VI) model when applied to a usably large workspace volume. A neural network is also used in tandem with the VI model to create a combined model. The output of the combined model is a VOI that can be used as an aid in a number of applications including robot path planning, entertainment, ubiquitous computing, and others.