Control and Optimization of Soft Exosuit to Improve the Efficiency of Human Walking

Abstract

Researchers have strived to develop lower limb wearable devices to improve the efficiency of human walking, however significant device inertia, kinematic constraints and the lack of optimal control strategies has limited performance when testing with human subjects. To alleviate these challenges, we developed the soft exosuit, a textile based wearable device that provides a conformal, unobtrusive and compliant means for interface with the human body. Unlike most rigid exoskeletons, the components of the soft exosuit on the lower extremity can be worn like light clothing. Combined with proximal mounted actuators, soft exosuit can provide forces that are in parallel to the lower limb muscles to improve the efficiency of walking. This thesis details the developments of soft exosuits, actuation platforms, real-time controllers and optimization methods. The working principle of the soft exosuits and the development of multi-joint actuation platforms that can provide real-time controlled assistance are presented. Two types of exosuit controllers, including iterative force-based position control and switching admittance-position control are designed and evaluated. The iterative force-based position control can accurately detect the hip extension onset timing and track peak timing and peak magnitude of the assistive profile within 1% of error. The switching admittance-position controller can track the whole shape of assistive profile with a root-mean-square error of 3.4%. For the exploration of human-robot interaction, human subject studies evaluating the effects of multi-joint assistance and timing of hip extension assistance are conducted. In the multi-joint assistance study with subjects carrying a 23.8 kg backpack on a treadmill of 1.25 m/s, the single joint condition solely assisting hip extension with a peak force of 95 N achieves a metabolic reduction of 4.6%, while the multi-joint condition assisting both hip extension and ankle plantarflexion with peak forces of 95 N and 196 N achieves a metabolic reduction of 14.6% compared to unpowered condition. The result suggests the metabolic benefits can be added up by assisting multiple joints simultaneously. In the effect of the timing of hip extension study, four different assistive profiles with two sets of onset and peak timings are evaluated with subjects carrying a 23 kg backpack on a treadmill of 1.5 m/s. The result demonstrates the actuation timing can affect the delivered mechanical power, biological joint power and metabolic performance. The assistive profile with an early onset and late peak timing delivers the most of mechanical power and achieves the highest metabolic reduction of 8.5% compared to the unpowered condition. Finally, the development and evaluation of human-in-the-loop Bayesian optimization for identifying the optimal control parameters is presented. The optimization method is first evaluated by finding subject’s optimal step frequency. It is found that the single-parameter optimization converges to the optimal step frequency in half the time of the established gradient descent method and significantly reduces the required total energy expenditure for the overall experiment protocol. Then, the multi-parameter Bayesian optimization is evaluated by optimizing the peak and offset timing of hip extension assistance with a soft exosuit. In this study, optimal timings are found over an average of 21.4 min and achieved 17.4% metabolic reduction compared to no-suit condition, which represented an improvement of more than 60% on metabolic reduction compared with state-of-the-art devices that only assist hip extension. The result provides evidence for participant-specific metabolic distributions with respect to peak and offset timing and metabolic landscapes, lending support to the hypothesis that individualized control strategies can offer substantial benefits over fixed control strategies. Overall, this thesis presents how to design and control soft exosuit system and lends insight into the human-robot interaction including the use of human-in-the-loop optimization to maximize the efficiency of human walking. I hope this work contributes to a shift towards individualization of future wearable devices.