Thesis title: Towards Advancements in Gait Assistance: Development and Testing of a Novel Two Degrees of Freedom AI Controlled Ankle Exoskeleton
Gait disorders, affecting millions globally, stem from neurological, orthopedic, and medical
conditions, with age as a significant risk factor. Stroke, a leading cause of adult disability,
often results in lower limb impairments that impact gait symmetry, stride, and efficiency,
frequently causing drop foot and push-off deficiencies. Such disorders limit daily activities,
increase health risks, and necessitate assistance, especially among older adults.
Rehabilitation is essential for stroke recovery; however, it remains time-consuming and costly.
Robotic systems, such as exoskeletons, present promising advancements in mobility assistance
and rehabilitation by addressing specific impairments. Consequently, joint-targeted
devices are emerging as specialized tools for both rehabilitation and assistance.
The complexity of the ankle joint poses significant challenges, often resulting in kinematic
oversimplifications that limit its physiological functionality. In response, we developed a
novel two-degree-of-freedom (DoF) ankle exoskeleton that provides active assistance for dorsiflexion
and plantarflexion while allowing passive inversion and eversion through a custom
joint mechanism. Three joint configurations were initially designed to evaluate how different
kinematic setups influence force exchange at the user-device interface. An experimental
trial with ten healthy subjects demonstrated significant differences in maximum pressure
variation across the gait cycle among the joint configurations.
An AI-based control algorithm was also developed, employing a convolutional recurrent
neural network (CRNN) to estimate the continuous gait phase (GP) based on signals from
a single inertial measurement unit (IMU). The neural model was validated in 14 healthy
subjects and compared with a previously validated algorithm in detecting four main gait
events: heel strike, mid stance, toe off, and mid swing. The trained model for continuous
GP estimation was then used to compute control torque, with the exoskeleton programmed
to assist with 12% of the user’s body weight in Newton meters (Nm).
The device’s performance was evaluated with twelve healthy subjects. Lower limb kinematics
were captured using an IMU, while muscle activity from twelve muscles in the right leg
was recorded via surface electromyography (EMG) during a treadmill walking task at three
different speeds. The study assessed kinematic impact using statistical parametric mapping
(SPM) to compare joint movements with and without the exoskeleton. Additionally, EMG
curve area variations were analyzed to evaluate changes in muscle activation.
The results demonstrated that the exoskeleton significantly impacted joint kinematics during
the push-off phase. Specifically, the ankle flexion angle increased during terminal swing,
indicating enhanced dorsiflexion provided by the device. EMG analysis revealed a significant
reduction in the activation of the soleus and medial gastrocnemius muscles during push-off,
while an increase in activation was observed in the biceps femoris and tibialis anterior during
the same phase.