SUMMARY

Slip incidents are some of the most common causes of workplace injuries annually in the US, as well as one of the most common causes of elderly falls. Due to the dangerous nature of slips, extensive research has been conducted into the causes and biomechanics of slips. Research has also delved into assisting with slip recovery, with one of the avenues pursued being the development of exoskeleton controllers for balance recovery. To run any of these controllers, a detection method was needed to indicate a slip has occurred before the controller could take effect. However, most of the research into slip detection has focused on detection accuracy, without reporting a detection time. Those that have reported detection time either had long detection times, or, if they were fast to detect slip, did not report an accuracy. The lack of confirmed fast and accurate detection methods has left a gap in the field of slip detection. Thus, this thesis will be focused on the development and validation of a machine learning algorithm to detect the onset of early and late slips in a person using only the sensors from a 1DOF hip exoskeleton both accurately and quickly. In pursuit of this goal, this thesis will cover 3 main sections: the development of a 1DOF hip exoskeleton for walking assistance, the development of a protocol to simulate both early and late slips at a range of magnitudes on a treadmill for data collection, and the optimization of a variety machine learning classification algorithms to try and create a quick and accurate slip detector.