Robust Adaptive Neural Force Control of a SEA-Based Knee Exoskeleton With Low Impedance

Various series elastic actuator (SEA)-based knee exoskeletons have been developed to assist disabled and able-bodied people during walking due to the advantages of SEA in force control, such as high power density, large force actuation, and low output impedance. However, ensuring efficient and accurate assistance is still challenging due to factors, such as the increased energy consumption of the unpowered proximal joint caused by the use of relatively heavy SEA and reduced control performance caused by the unknown model of the human leg and fast disturbances from humans and the environment. This article presents a novel knee exoskeleton design combining a linkage mechanism with a linear SEA to shift the motor and gearbox to the hip joint to reduce the weight of the knee joint, leading to much lower inertia regarding the hip joint and, consequently, lower energy consumption. Moreover, we propose a robust adaptive sliding mode-based force control (RASMC) method, which efficiently integrates a neural network and a second-order disturbance observer to estimate the human–exoskeleton model and transient disturbances for compensation synchronously. The stability of the proposed RASMC method is theoretically analyzed and proved based on the Lyapunov theorem. The force control performance of the proposed method is evaluated via experiments by comparing it to conventional control methods. The experimental results demonstrate that the RASMC approach can ensure high force tracking accuracy and strong robustness to model changes and disturbances.