Parameter Optimization for Neural Oscillators Applied to Trajectory Generation of an Exoskeleton for Lower Limbs

This work presents an optimization system developed to find optimal parameters for neural oscillators used for trajectory generation of an exoskeleton for lower limbs. An exoskeleton can be considered as a biped robot, with a given force interaction between the user and the robot. Since biped robots must present cyclical joint trajectories during the walking, neural oscillators are being used as trajectory generators for these systems. The neural oscillators used in this work are based on Matsuoka oscillators, consisting of two mutual inhibitory neurons which are able to produce a cyclical output. Each neuron is modeled by two differential equations whose parameters are difficult to be set for a given desired output. In this way, it was developed an optimization system to find the better values to the neural oscillator parameters given a predefined desired joint trajectory. This optimization system works to minimizing the error between the trajectory generated by the oscillator and the desired trajectory, regarding the robot dynamics. The advantage of using oscillators is justified because the trajectories can be generated in a on-line process, with a less time-consuming with relation to analytical methods. The results show that the proposed optimization system and the trajectory generator using neural oscillators can be applied in an adaptive model that include interaction forces between the user and the robot, so changing the trajectory according to the user intention.