Grimmer Martin, Zeiss Julian, Weigand Florian, Zhao Guoping
Lauflabor Locomotion Laboratory, Department of Human Sciences, Institute of Sports Science, Technical University of Darmstadt, Darmstadt, Germany.
Department of Electrical Engineering and Information Technology, Institute of Automatic Control and Mechatronics, Technical University of Darmstadt, Darmstadt, Germany.
Front Neurorobot. 2022 Oct 6;16:948093. doi: 10.3389/fnbot.2022.948093. eCollection 2022.
Human-in-the-loop (HITL) optimization with metabolic cost feedback has been proposed to reduce walking effort with wearable robotics. This study investigates if lower limb surface electromyography (EMG) could be an alternative feedback variable to overcome time-intensive metabolic cost based exploration. For application, it should be possible to distinguish conditions with different walking efforts based on the EMG. To obtain such EMG data, a laboratory experiment was designed to elicit changes in the effort by loading and unloading pairs of weights (in total 2, 4, and 8 kg) in three randomized weight sessions for 13 subjects during treadmill walking. EMG of seven lower limb muscles was recorded for both limbs. Mean absolute values of each stride prior to and following weight loading and unloading were used to determine the detection rate (100% if every loading and unloading is detected accordingly) for changing between loaded and unloaded conditions. We assessed the use of multiple consecutive strides and the combination of muscles to improve the detection rate and estimated the related acquisition times of diminishing returns. To conclude on possible limitations of EMG for HITL optimization, EMG drift was evaluated during the Warmup and the experiment. Detection rates highly increased for the combination of multiple consecutive strides and the combination of multiple muscles. EMG drift was largest during Warmup and at the beginning of each weight session. The results suggest using EMG feedback of multiple involved muscles and from at least 10 consecutive strides (5.5 s) to benefit from the increases in detection rate in HITL optimization. In combination with up to 20 excluded acclimatization strides, after changing the assistance condition, we advise exploring about 16.5 s of walking to obtain reliable EMG-based feedback. To minimize the negative impact of EMG drift on the detection rate, at least 6 min of Warmup should be performed and breaks during the optimization should be avoided. Future studies should investigate additional feedback variables based on EMG, methods to reduce their variability and drift, and should apply the outcomes in HITL optimization with lower limb wearable robots.
有人提出通过代谢成本反馈进行人工介入优化，以减少可穿戴机器人辅助下的行走能耗。本研究探讨下肢表面肌电图（EMG）是否可作为一种替代反馈变量，以克服基于代谢成本的探索耗时问题。在实际应用中，应能够根据肌电图区分不同行走能耗的情况。为获取此类肌电图数据，设计了一项实验室实验，让13名受试者在跑步机上行走时，在三个随机的负重阶段加载和卸载成对的重物（总计2、4和8千克），以引发用力程度的变化。记录了双下肢七块下肢肌肉的肌电图。使用负重加载和卸载前后每一步的平均绝对值来确定在负重和非负重状态之间变化的检测率（如果每次加载和卸载都能相应检测到，则为100%）。我们评估了使用多个连续步幅和肌肉组合来提高检测率，并估计了收益递减的相关采集时间。为了总结肌电图在人工介入优化中的可能局限性，在热身阶段和实验过程中对肌电图漂移进行了评估。多个连续步幅的组合和多块肌肉的组合使检测率大幅提高。肌电图漂移在热身阶段和每个负重阶段开始时最大。结果表明，在人工介入优化中，使用多块相关肌肉的肌电图反馈以及至少10个连续步幅（5.5秒），可受益于检测率的提高。结合最多20个排除的适应步幅，在改变辅助条件后，建议进行约16.5秒的行走以获得可靠的基于肌电图的反馈。为了最小化肌电图漂移对检测率的负面影响，应至少进行6分钟的热身，并避免在优化过程中出现中断。未来的研究应基于肌电图研究其他反馈变量、降低其变异性和漂移的方法，并将结果应用于下肢可穿戴机器人的人工介入优化中。
