Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Electrical and Computer Engineering


Dr. Kenneth A. McIsaac


In recent years, numerous researchers have been working towards adapting technology developed for robotic control to use in the creation of high-technology assistive devices for the visually impaired. These types of devices have been proven to help visually impaired people live with a greater degree of confidence and independence. However, most prior work has focused primarily on a single problem from mobile robotics, namely navigation in an unknown environment. In this work we address the issue of the design and performance of an assistive device application to aid the visually-impaired with a guided reaching task. The device follows an eye-in-hand, IBLM visual servoing configuration with a single camera and vibrotactile feedback to the user to direct guided tracking during the reaching task.

We present a model for the system that employs a hybrid control scheme based on a Discrete Event System (DES) approach. This approach avoids significant problems inherent in the competing classical control or conventional visual servoing models for upper limb movement found in the literature. The proposed hybrid model parameterizes the partitioning of the image state-space that produces a variable size targeting window for compensatory tracking in the reaching task. The partitioning is created through the positioning of hypersurface boundaries within the state space, which when crossed trigger events that cause DES-controller state transition that enable differing control laws. A set of metrics encompassing, accuracy ($D$), precision ($\theta_{e}$), and overall tracking performance ($\psi$) are also proposed to quantity system performance so that the effect of parameter variations and alternate controller configurations can be compared.

To this end, a prototype called \texttt{aiReach} was constructed and experiments were conducted testing the functional use of the system and other supporting aspects of the system behaviour using participant volunteers. Results are presented validating the system design and demonstrating effective use of a two parameter partitioning scheme that utilizes a targeting window with additional hysteresis region to filtering perturbations due to natural proprioceptive limitations for precise control of upper limb movement. Results from the experiments show that accuracy performance increased with the use of the dual parameter hysteresis target window model ($0.91 \leq D \leq 1$, $\mu(D)=0.9644$, $\sigma(D)=0.0172$) over the single parameter fixed window model ($0.82 \leq D \leq 0.98$, $\mu(D)=0.9205$, $\sigma(D)=0.0297$) while the precision metric, $\theta_{e}$, remained relatively unchanged. In addition, the overall tracking performance metric produces scores which correctly rank the performance of the guided reaching tasks form most difficult to easiest.