Optimized hip-knee-ankle exoskeleton assistance reduces the metabolic cost of walking with worn loads

Background Load carriage is common in a wide range of professions, but prolonged load carriage is associated with increased fatigue and overuse injuries. Exoskeletons could improve the quality of life of these professionals by reducing metabolic cost to combat fatigue and reducing muscle activity to prevent injuries. Current exoskeletons have reduced the metabolic cost of loaded walking by up to 22% relative to walking in the device with no assistance when assisting one or two joints. Greater metabolic reductions may be possible with optimized assistance of the entire leg. Methods We used human-in the-loop optimization to optimize hip-knee-ankle exoskeleton assistance with no additional load, a light load (15% of body weight), and a heavy load (30% of body weight) for three participants. All loads were applied through a weight vest with an attached waist belt. We measured metabolic cost, exoskeleton assistance, kinematics, and muscle activity. We performed Friedman’s tests to analyze trends across worn loads and paired t-tests to determine whether changes from the unassisted conditions to the assisted conditions were significant. Results Exoskeleton assistance reduced the metabolic cost of walking relative to walking in the device without assistance for all tested conditions. Exoskeleton assistance reduced the metabolic cost of walking by 48% with no load (p = 0.05), 41% with the light load (p = 0.01), and 43% with the heavy load (p = 0.04). The smaller metabolic reduction with the light load may be due to insufficient participant training or lack of optimizer convergence. The total applied positive power was similar for all tested conditions, and the positive knee power decreased slightly as load increased. Optimized torque timing parameters were consistent across participants and load conditions while optimized magnitude parameters varied. Conclusions Whole-leg exoskeleton assistance can reduce the metabolic cost of walking while carrying a range of loads. The consistent optimized timing parameters across participants and conditions suggest that metabolic cost reductions are sensitive to torque timing. The variable torque magnitude parameters could imply that torque magnitude should be customized to the individual, or that there is a range of useful torque magnitudes. Future work should test whether applying the load to the exoskeleton rather than the person’s torso results in larger benefits. Supplementary Information The online version contains supplementary material available at 10.1186/s12984-021-00955-8.


1.Participant information, validation protocol
For this study, we revalidated the no load condition for participants 1 and 2 to allow for a more direct comparison between conditions and participants. The original validation experiments occurred over 7 months earlier and before the implementation of Covid-19 masking protocols. The re-validation sessions accounted for any physical changes the participants underwent over the 7 month period, such as overall fitness or weight, as well as the protocol change to include the paper or cloth masks. We did not revalidate the no load condition for participant 3 because they completed the no load validation experiment less than a month before beginning this protocol when masking protocols were already in place. All three participants underwent identical laboratory testing procedures.   Muscle activity averaged over a stride for participant 3. Increased semitendinosus and bicep femoris activity near 50% of stride is due to the sensor interacting with a thigh strap. Gluteus maximus activity with the light load was impacted by interactions of the weight vest and exoskeleton waist strap. Increased gluteus maximus activity for the heavy load near 60% of stride was due to strap interactions with the sensor.

Minimum value subtraction
To process the muscle activity data, the data was bandpass filtered at 40 and 450 Hz, rectified, then low pass filtered at 10 Hz. We also subtracted the minimum signal value because high frequency noise produced a constant offset that we were unable to remove through filtering. The offset implied constant muscle activation specifically during times of rest. For example, soleus electromyography signals were non-zero during swing when it is typically quiescent. This subtraction impacted the RMS activity. In some instances, subtracting the offset produced a smaller RMS value than when the offset was not subtracted, while in other cases, the subtraction resulted in larger RMS values. Muscle activity results are plotted below with and without the minimum value subtracted.  (orange) conditions. The bottom row shows the RMS of the muscle activity with assistance for all load conditions. The RMS of the unassisted muscle activity is shown with the gray line (dashed). Muscle activity was normalized to the unassisted activity resulting in a peak value of 1 for unassisted walking at all loads.

Kinematic results
Ground reaction forces Figure 6. Ground reaction forces in the x and y directions for the left and right belts. No load (blue), light load (green) and heavy load (orange) results are shown. Both x and y ground reaction forces increased with load. Exoskeleton assistance typically decreased the magnitude of the first peak and increased the magnitude of the second.
Stride frequency  The hip profile was defined by 8 parameters (Fig. 2 Hips). These parameters defined the rise time, peak time and peak magnitude of hip extension and the peak time, peak magnitude and fall time of hip flexion. A period of no torque was prescribed between extension and flexion periods and was defined by the mid-point timing and duration.
The knee profile was defined by 10 parameters (Fig. 2 Knees). It consisted of a virtual spring during stance, time based flexion torque near toe off, and a virtual damper during swing. The virtual spring was defined by the stiffness, onset time and offset time. The spring torque was the stiffness multiplied by the knee joint angle, which was set to zero when the knee was straight. Knee flexion torque near toe off was defined by the peak time, peak magnitude, rise time and fall time. The virtual damper during swing was parameterized with a damping coefficient, onset time and offset time, similar to the parameterization of the virtual spring.
The ankle profile was defined by 4 parameters similar to knee flexion near toe-off (Fig. 2 Ankles). These defined the peak time, peak magnitude, rise time, and fall time. Ankle torque was set to zero at 65% of stride at the latest to avoid torque application during swing. This constraint could shorten the fall time, for example a peak time of 55% of stride would result in a maximum fall time of 10% of stride.

Parameter ranges and optimized values Hip parameter ranges and optimized values
The hip profile was defined by 8 parameters. It applied hip extension torque through heel strike, so the stride timer began at 84% of stride to avoid discontinuities in the desired profile at heel strike. To convert the hip timing parameters to be based off of heel strike, subtract 16% from the current value. Table 9. Hip parameter ranges for all speeds and initial values for the no-load condition. The light load condition was initialized with the optimized values from the no-load condition, and the heavy load condition was initialized with the optimized values from the light load condition.

Knee parameter ranges and optimized values
The knee profile was defined by 10 parameters. The stride time is 0% at heel strike and 100% at the following heel strike of that leg. The knee profile has two state based periods, a virtual spring during stance and a virtual damper during late swing. These two periods were defined by the onset and offset timing of the periods and by the stiffness or damping constant. During the period of virtual spring torque, if the knee joint angle went to 0 before the end of the period, the exoskeleton stopped applying torque for participant comfort.

Ankle parameter ranges and optimized values
The ankle profile was defined by 4 parameters. The stride time is 0% at heel strike and 100% at the following heel strike of that leg. *Torque was limited to be applied no later than 65% of stride, so, for example, if peak time was at its latest allowed value (55% of stride), fall time was limited to be 10% of stride.  Table 15. Root mean square torque tracking error for the no-load, light load and heavy load conditions.The error is reported in Nm and as percent of the maximum torque. Knee assistance typically resulted in more torque tracking error. The state based periods have step changes in desired torque and change on a step to step basis. Iterative learning, part of the control structure, learns the torque tracking error over time, so it is slightly less effective when the desired torque changes on each step than it is with consistent desired torque. 8. Impact of the cloth mask Figure 8. Metabolic impact of the cloth mask under the metabolic mask. Participant 1 measured their metabolic cost of quiet standing and walking on the treadmill for 6 minutes each. All measurements were collected in the same session. They measured their metabolic cost with no cloth mask (blue) and with a cloth mask under the metabolics mask (red). The average of the last 3 minutes of walking is shown with the solid line, and the dashed lines show the range for one standard deviation. The participant was the only person in the lab space, so they did not put others at risk by not wearing a mask. The cloth mask lowered the metabolic cost by 0.33 W/kg for the standing condition and by 0.68 W/kg for the walking condition.