US ARMY DEVCOM Army Research Laboratory, United States
Introduction:: The battlefield of the future will likely employ adaptive intelligent agent teammates for the purpose of easing physical and cognitive burden and improving Soldier operational effectiveness. Adaptive exoskeletons are a candidate technology which may be useful in alleviating physical burden and augmenting performance, but their successful implementation depends upon understanding how humans and intelligent systems mutually adapt to each other to achieve optimal performance.
We conducted a study to investigate neural and biomechanical changes that occur as novice users learn to cooperate with an intelligent plantarflexion assistance exoskeleton (exo). Initial analyses comparing muscle activation before (PRE) and after (POST) about 45 minutes of walking have revealed that muscle activation was reduced during the POST trials for all muscles involved in the walking task, but plantarflexor muscles also exhibited reduced activation in response to the exo assistance [1,2]. Here, we investigate the evolution of these changes over time by examining incremental observations collected as the exo was powered OFF and ON during the 45 minutes of walking. We examine the trend of the muscle activation reduction to determine if adaptation plateaued during the training, and we compare linear regression models to determine if differences exist in the rate of adaptation due to actuation state of the exo (OFF vs ON). We hypothesized that assistance ON would result in greater adaptation rates for the plantarflexor muscles compared to OFF, and we hypothesized that the rates of adaptation would be equivalent between actuation states for more proximal muscles.
Materials and Methods:: Twenty-two healthy, active subjects (15M/7F) with no exo experience and with no musculoskeletal injuries were recruited for participation. Participants were instrumented with 128-channel electroencephalography (EEG), 10-channel electromyography (EMG), and optical motion capture markers on the head, trunk, and lower extremities. Bipolar surface EMG electrodes (Noraxon USA, Inc., Scottsdale, AZ) were placed bilaterally over the rectus femoris (RF), biceps femoris (BF), tibialis anterior (TA), lateral gastrocnemius (GS), and soleus (Sol) muscles. Participants also donned bilateral powered ankle exos (Dephy, Inc., Maynard, MA) fitted to the nearest whole shoe size. Participants completed initial walking trials (PRE) of 5 minutes duration with the exo powered OFF and ON. Participants then completed approximately 45 minutes of walking during which the exo power toggled between OFF and ON states 30 times without warning. Finally, participants completed a final 5 minutes of walking for both the OFF and ON conditions.
EEG, EMG, and kinematic data were recorded synchronously throughout all trials. Differences between the PRE and POST trials have previously been reported [1,2]. Here, we report the EMG-based changes during the 45-min power cycling. EMG data for each muscle was filtered using a 4th-order, zero-lag bandpass filter (10-450Hz), demeaned, and rectified. Each stride was normalized to 101 data points from heel strike to heel strike, and signal amplitudes were normalized to the average maximum amplitude during the PRE OFF trial. Applying trapezoidal integration to these time- and magnitude-normalized EMG signals, a unitless mean total muscle activation was calculated for each muscle for each of 30 exposures.
Results, Conclusions, and Discussions:: Results
Total activation was averaged for each muscle across both left and right legs and is presented as a function of observation number (Figure 1). The present analyses include a subset of 6 participants, but final analyses will include data from all participants.
Linear regression analyses were conducted using total activation as the dependent variable and observation, condition, and their interaction as independent variables. Models were significant for each muscle. For RF and TA muscles, only observation number had a significant effect on total activation (F3,56=56.63, p< 0.0001, t=-8.16, p< 0.0001; F3,56=34.87, p< 0.0001, t=-6.16, p< 0.0001). For GS (F3,56=64.88, p< 0.0001) and Sol (F3,56=112.8, p< 0.0001), both observation (t=-6.7, p< 0.0001; t=-8.63, p< 0.0001) and condition (t=-4.28, p< 0.0001; t=-6.65, p< 0.0001) were significant contributors to total activation. The BF model (F3,56=68.91, p< 0.0001) revealed a significant effect of observation (t=-5.70, p< 0.0001) and also a significant interaction of observation and condition (t=-3.91, p< 0.0001).
Conclusions
Our hypotheses were partially supported by our analyses. As anticipated, the more proximal RF and TA muscles exhibited similar rates of adaptation for both OFF and ON conditions, reducing total activation over time. Contrary to our hypotheses, however, adaptation rates were similar between conditions for the plantarflexor muscles, but these muscles exhibited a more immediate and constant effect of the exo assistance while the BF exhibited different adaptation rates between conditions.
Discussions
Initial analyses [1,2] revealed that muscle activation was reduced as subjects practiced with the exo regardless of whether it was powered or unpowered. This is indicative of subjects learning to accept the exo itself regardless of activation state. The negative slopes for all conditions indicate that adaptation was likely incomplete at the conclusion of the paradigm. The condition effect for the GS and Sol demonstrates that muscles most directly augmented by exo assistance will exhibit a more immediate effect of the exo. The state-dependent differences in BF adaption rate suggest that some muscles which peripherally support those directly augmented may undergo adaption over a longer time scale as new gait patterns are adopted.
Acknowledgements (Optional): : The authors would like to thank Ms. Anna K McGough and Mr. Christian Poindexter for their assistance with data collection.
References (Optional): : [1] Haynes CA, Bradford JC, Song S. (2022) Using muscle synergy analysis to investigate exoskeleton adaptation. Proceedings of the North American Congress on Biomechanics, Ottawa, Ontario, Canada, Aug 21-25, 2022.
[2] Song S, Haynes CA, Bradford JC. (Preprint) Human cortical, muscular, and kinematic gait adaptation with novel use of an ankle exoskeleton. Under Review by Scientific Reports. https://doi.org/10.21203/rs.3.rs-2675191/v1