Real-Time Experimental Validation of an Adaptive Gait Event Detection Algorithm

The accurate detection of gait events, such as heel strike (HS) and toe off (TO), is critical for the implementation of many lower limb exoskeleton control strategies. While underfoot force sensors are commonly used, their limitations have inspired the development of algorithms to detect these events from kinematic information obtained from inertial measurement units (IMUs) placed on limb segments and encoders at the joints. This work presents an adaptive, user-independent approach to gait event detection, using kinematic information from the embedded sensors of an exoskeleton. These included hip and knee joint encoders and a thigh-worn IMU; additionally, event detection only considered ipsilateral data. The algorithm was evaluated in real time with seven healthy subjects walking in the Indego Explorer exoskeleton on an instrumented treadmill at varying speeds and slopes. The experiment yielded mean gait phase errors of 0.38% $\pm ~1.82$ % for HS and 0.54% $\pm ~1.14$ % for TO at an accuracy of 99.93% over a total of 4104 steps. The results of this validation suggest that the presented algorithm is a viable solution to inertial-based gait event detection for the application of lower limb exoskeleton control. Future directions for this work include evaluating the algorithm for overground walking and for a controller providing gait assistance.