Efficient Dynamics Modeling of Industrial Robots in Encoded Monoid Space

Abstract

For conventional linear-in-parameter (LIP) dynamic model of industrial robots, redundant terms in multivariate polynomials (MVPs) are strongly coupled and more difficult to eliminate than the redundant parameters. In this work, by analyzing the mapping from linearized chain kinematics to Lagrange dynamics, the linear-in-multivariate-polynomial (LI-MVP) model is formed. Subsequently, a bilinearized dynamics is derived in both LI-MVP and LIP formulations, so that redundant MVPs and inertial parameters are concurrently eliminated. In addition, all the MVPs are encoded into a numeric matrix within a monoid, where a binary operation is defined to replace symbolic Kronecker products for efficient derivation of LI-MVP model. Eventually, by rearranging the encoded matrix with respect to total degrees of variate, the symbolic LIP model is directly restored with decoded MVPs in Horner forms, which can further reduce the number of multiplications during torque computation. Simulations and experiments on serial industrial robots in both model derivation and torque calculation demonstrate the effectiveness of proposed methods.