Online Jumping Motion Generation via Model Predictive Control

Abstract

Legged robots can move on various grounds via walking; however, they have difficulty in moving fast. To overcome this limitation, some legged robots have been studying dynamic motions such as jumping and running. In this article, we present a strategy for generating an optimal jumping motion for legged robots in real time. For convenience, we divided the proposed jumping trajectory into vertical and horizontal directions. The vertical motion trajectory ensures a continuous center of mass position, speed, and acceleration, as well as minimizes the maximum torque and maximum speed of the joints; we generated it via a nonlinear optimal process. Besides, for the horizontal motion, we proposed a novel model predictive control using a height varying inverted pendulum model. In the proposed method, the zero moment point is placed in the support polygon for stable jumping; also, the robot does not slip on the ground and takes off at a desired velocity. We verified the jumping motion using RoK-3, a biped robot developed for the experiment.