Human Center of Mass Trajectory Models Using Nonlinear Model Predictive Control

Abstract

Abstract In this paper we describe a nonlinear model predictive control (NMPC) scheme to model human Center of Mass (CoM) trajectories in an environment with an obstacle. This NMPC model can be used onboard a robot for planning its motion in a way humans do, or can in addition be used by the robot to predict a probable future CoM trajectory of a human participant. Both features support human-robot interaction. The focus of this work is on the derivation of appropriate parameters of the NMPC and a suitable geometrical model of a safety zone around the obstacle. To this end, we have conducted experiments with human participants and have computed the resulting CoM trajectories. On this basis we have derived suited NMPC parameters to imitate the human CoM trajectory planning behavior with a high precision. With models tailored to each experiment, the NMPC approximates the human motion with mean distances of less than 5cm between modelled and real CoM trajectory. With a generalized model, which is identical for all experiments, the mean distance per experiment is less than 12cm.