Dynamical Movement Primitives of Crossing over Obstacles for a Lower-Limb Prosthesis

Abstract

This paper proposes a trajectory generation of prosthesis to help amputee crossing over obstacles. Dynamical movement primitives (DMPs) are popular methods of reproducing trajectory for learning control. In the basic of DMPs, a novel term of obstacle is added to generate the trajectory in real-time. This term includes multiple point obstacle sources to reflect spatial size of obstacles. Each point obstacle is used to calculate direction relative to current position. Moreover, the direction vector and velocity vector are considered into term of obstacle. Besides, DMPs can generate effective trajectory through obstacle term parameter adjusting regardless of the obstacles in the front, middle or behind position in a step. Therefore, our method solves the inability crossing over obstacles under some scenarios and helps the prosthesis adapt to the various environment. Gaussian mixture regression is used to conjunction with DMPs for movement representation, which can reproduce new trajectory from multiple sets of original trajectories. Finally, simulations for DMPs were performed to demonstrate the flexibility of DMPs with obstacle term, which can generate desired trajectory of crossing over obstacles in various situations, indicating our method has a potential to facilitating prosthesis motion control.