Based on the results of the gait patterns, the PI will design appropriate exercises that target the specific gait impairment at a specific time period during the gait cycle. Exercises will utilize various modalities that are commonly used in clinical practice, including TheraBand, gym equipment, free weight. The training may target single-joint movement or targeting a movement pattern that involves multiple joint holistically. A body harness system will be used to protect the participant during the exercise training. The participant will be supervised by a research assistant who has rich experience in conducting exercise training on people with disability.

Participants in the standard exercise group will undergo a standardised exercise programme consisting of five lower-limb exercises commonly employed in clinical settings and prescribed to patients as home exercises. They include (1) stepping up and down to strengthen the muscles of both legs and to improve control in shifting the center of gravity; (2) heel-raising exercises on an inclined wedge to strengthen both ankle plantarflexors; (3) assuming a semi-squatting position to improve lower-limb muscle endurance and proprioception in the knees and ankles; (4) standing on a dura disk to improve dynamic standing balance; (5) walking across obstacles to enhance anticipatory postural control.

The result from the Statistical Parametric Mapping 1D (SPM1D) analysis will be used to identify the magnitude of deviation from normal gait patterns in reference to healthy subjects for each lower limb joint. Thereby, the targeted action and joint position can be identified. Training will focus on the specific movement where the "steepest gradients" are observed.

The targeted exercises will be selected based on the five largest deviations observed during the swing or stance phase. Open-chain exercises will be applied to address deviations in the swing phase, while closed-chain exercises will be applied to address deviation in the stance phase. Each exercise will be performed for approximately 80-100 repetitions, with a total duration of 10 minutes per exercise, including rest periods.

Seventeen wearable sensors (Inertial Measurement Units, IMUs) were attached to the participants' bodies in accordance with the Xsens user manual to collect kinematic data. The extracted data included the pelvis tilt angle (°), oblique angle (°), and rotation angle (°); hip flexion angle (°) and abduction angle (°); knee bending angle (°); as well as ankle flexion angle (°) and abduction angle (°), which were used for further analysis. These data were converted into standardized gait cycles (101 time points) for statistical parametric mapping (SPM1D) analysis, and the deviation time from the normal reference was compared to determine whether there were changes in deviation timing before and after the treatment.

Twelve Electromyography (EMG) sensors were placed on the left and right sides of the gluteus maximus, gluteus medius, quadriceps, hamstrings, tibialis anterior, and gastrocnemius muscles to measure Maximum Voluntary Contraction (MVC). In addition, the timing of each muscle activiation during gait cycle will be extracted to compared before and after training. This includes identifying the onset (activation start) and offset (activation end) of each muscle relative to specific phases of the gait cycle, such as the stance and swing phases.