Adaptive Admittance Control of Human-Exoskeleton System Using RNN Optimization

For wearable exoskeletons, ensuring safety and stability while reducing the effort of human lower limbs during walking remains a challenge. The level of safety can be guaranteed by constraining the exoskeleton robot to never exhibit any erratic behavior. But beyond that, the cooperative motion generated in human-exoskeleton interactions should be truly intuitive and should minimize the limitations of human performance. Therefore, this paper presented a new adaptive admittance control law, which not only ensures the stability of the lower limb exoskeleton in the constrained motion but also provides a very harmonious human-exoskeleton interaction. The proposed approach is applied to the movement of the exoskeleton containing the knee joint in the swing phase. The input of the admittance control system is the human-exoskeleton interaction force measured by one single-dimensional force sensor. The performance of the method was theoretically analyzed and discussed. Experiments were conducted in Matlab simulation environment to verify the effectiveness of aid to the swinging leg. Compared with the fixed admittance control, the adaptive admittance control system optimized by a Recurrent Neural Network(RNN) greatly improves the level of interaction between the exoskeleton and the wearer.