Figure 1.
RunBot's basic operation involves phase switching of the legs triggered by contact signals from the feet.
(A) Photographs of RunBot's gait cycle and (B) the system used by RunBot to generate stepping.
Figure 2.
Set-up for the treadmill walking trials.
(A) The USB-DUX Sigma data acquisition device and EMG/FSR amplifier are worn in a waist bag around the subject's waist. Surface EMG electrodes are used to record the muscle activity during the treadmill walking. FSR insoles are placed in the subject's shoes and measure contact signals under different areas of the feet. (B) Position of the recorded muscles on the leg. TA = Tibialis Anterior, RF = Rectus Femoris, BF = Biceps Femoris and LG = lateral Gastrocnemius.
Figure 3.
EMG and FSR foot contact data from each subject was recorded over a range of different walking speeds.
The data could then be separated depending on the walking speed and compared to the activity recorded over the entire sequence (black dashed line in figure). To analyse the activity before and after heel contact, an event related average (ERA) was taken in a time period of one stride duration before and after the heel contact. The figure demonstrates the relationship between left leg smoothed and rectified EMG and heel contact information from one of the ten subjects during walking sequence 1 (25 steps per speed setting). Increasing walking speed increases the amplitude of the EMG signal, as described by [28]. TA = Tibialis Anterior, LG = Lateral Gastrocnemius, RF = Rectus Femoris and BF = Biceps Femoris.
Figure 4.
Photograph series representing one gait cycle during treadmill walking.
The series of frames corresponds to one stride from heel strike of the left leg (highlighted in white in the first and last frame) to the next heel strike of the same leg. The filter output using the transfer functions for each measured muscle of the left leg corresponding to the heel strike of the ipsilateral leg, found using the adaptive filtering, are shown alongside the images of one stride duration. (A) = , (B) =
, (C) =
and (D) =
, HS = Heel strike.
Figure 5.
Identifying features of the transfer function coefficients which correspond to muscle activity promoting knee and hip flexion/extension in human walking.
The transfer functions from adaptive filtering heel contact data from the contralateral and ipsilateral foot to the specific leg muscle ((A) , (B+C)
, (D)
, (E)
and (F)
) were used to identify the required features. These coefficients were then used in an FIR filter to control motors in RunBot's hip and knee using the sensory input of the contralateral or ipsilateral heel contact.
Table 1.
Relating the muscle transfer function to RunBot's motor control.
Table 2.
Results of the curve fitting for hip flexion/extension.
Table 3.
Results of the curve fitting for knee flexion/extension.
Table 4.
Sets of transfer functions applied to RunBot's control system.
Figure 6.
Plots of the different transfer functions tested with RunBot which produced a stable gait.
The number of samples for the hip motors was set to 200 (1000 ms). This is the same frequency used during the normal operation of RunBot II. Knee flexion/extension was set to 100 samples or 500 ms. (A) Represents the transfer function coefficients from the curve fitting for the hip flexion. Hip flexion of the leg is triggered by the contralateral heel strike. (B) Hip extension is triggered by the ipsilateral heel strike. (C) Knee flexion of the leg is triggered by the contralateral heel strike and knee extension (D) is triggered by the anterior extreme angle (AEA) of the hip to drive knee extension at terminal swing.
Table 5.
Comparison of function characteristics.
Figure 7.
Photographs of one RunBot stride duration.
The series of frames corresponds to one stride recorded after applying transfer functions found from the human study. The time interval between each adjacent frame is 60 ms. Markers were attached to RunBot's right leg for video tacking of the joints for calculation of kinematic data. Heel contact triggers the stance phase of ipsilateral leg and the swing phase of the contralateral leg. Leg extension during terminal swing is triggered by the threshold value for the hip anterior extreme angle (AEA) being reached during hip flexion.
Figure 8.
Box plots comparing RunBot's stride length (A), stride duration (B) and walking speed (C).
Using the transfer function sets which produced stable walking (n = 10). A box plot comparing the relative walking speed of RunBot using each of the transfer function sets compared to the average relative walking speed of the human test subjects is also provided (D). Relative walking speed of leg-length/s is calculated as the scaled walking speed to leg length where RunBot's leg length is 0.3 m.
Figure 9.
Comparison of knee flexion/extension angle of RunBot using transfer function sets which produced a stable gait.
The time is normalised to percent of stride, the mean and standard deviation was calculated from the number of strides recorded from the video tracking. The mean percent of stride when the contralateral heel strike was recorded is also shown as a line with the standard deviation highlighted.
Figure 10.
Phase plots of knee angular velocity versus angular position.
Knee angle was calculated from markers positioned on the knee joint and video camera tracking over ten complete rotations of the circular path. The plots show the limit cycles in the phase plane and demonstrate the robustness of the reflexive control system, as even when there is a disturbance to the gait cycle there is a rapid convergence to the limit cycle in only a few steps.