Dynamic modeling of robot based on neural network with incomplete state observations

Abstract

This paper presents a novel dynamic modeling method of robot system using a recurrent neural network (RNN) with incomplete state variables observation. A dynamic model of a 2-DOF articulated robot is discussed, and the corresponding training method is deduced based on the back propagation through time (BPTT) algorithm. The effectiveness of this process is verified by simulation. The results show that the observed state variables are regressed, and the unobserved state variables are estimated.