UNS: Optimal Adaptive Control Methods for a Hybrid Exoskeleton

点击翻译按钮获取中文摘要

Restoration of walking and standing function is one of the most desired interventions that would improve quality of life of persons with paraplegia. Evidence supports that the users of walking devices experience fewer secondary medical complications than wheelchair users. To restore walking and standing function, a hybrid exoskeleton that combines functional electrical stimulation (FES) with a powered exoskeleton can be more advantageous than an FES-based walking system or a powered exoskeleton alone. Little research has gone into the design and evaluation of control methods for a hybrid exoskeleton. Using FES and an electric motor, the proposed research will investigate an automatic control method to produce shared or cooperative force at a limb joint. The proposed algorithm will be able to maintain cooperation between FES and the electric motor even when FES-induced muscle fatigue sets in. Advances in shared control imply that smaller and light weight exoskeletons can be designed because FES can be additionally used to exploit a user's inherent muscle power. The use of FES in the hybrid neuroprosthesis will also provide therapeutic benefits; e.g., application of FES improves cardiovascular fitness, motor-skill relearning, and increased muscle mass and fatigue resistance. Thus, the proposed study will provide substantial benefits to society by improving quality of life of individuals with mobility impairments and increasing their community participation.The proposed research will investigate cutting-edge control methods that are based on reinforcement learning (RL) principles for simultaneously controlling FES and an electric motor. The current ad hoc techniques used to control both FES and an active orthosis together in a hybrid device do not necessarily adapt with muscle fatigue - a major limiting factor in an FES-based technology. Moreover, the dissimilar dynamics of FES and the electric motor can cause instability during walking, which can potentially lead to injury due to falling. The new RL control techniques will compute approximate optimal solutions in real-time and will be applicable to dynamic systems that are driven by multiple actuators with dissimilar dynamics. The specific aims of the proposal are to investigate and evaluate RL-based actor-critic control methods on a hybrid leg extension machine and a hybrid walking device. The proposed experiments will measure any reduction in the overall power requirement of the hybrid system and its dependence on the muscle fatigue. The research will also result in a phenomenological model of common peroneal nerve (CPN) stimulation, which is used to elicit hip flexion during walking. This model will build an understanding on habituation and elicitation characteristics of the CPN stimulation. In collaboration with a physiatrist, experiments will be conducted on able-bodied subjects and a participant with spinal cord injury.