Online Noise-Estimation-based Neighbor Selection for Multi-Manipulator Systems

Abstract

Abstract In this paper, a novel online neighbor selection policy is proposed in the control of nonlinear networked multi-manipulator systems where manipulators’ joints’ signals are subject to varying noise levels. By addressing the issue in many conventional control methods of multi-agent systems (MASs) where all available neighbor signals are used without evaluating the quality of the information, efforts of this paper seek to improve the overall tracking performance by actively selecting neighbor feedback signals in the robust non-singular terminal sliding mode (NTSM) control. A fast neighbor selection scheme is presented by incorporating an online noise covariance estimation into a nonlinear continuous-discrete unscented Kalman filter (CD-UKF). A selection index vector is recursively updated by the estimated noise covariance matrix for the control design. Simulation results of a group of six degrees of freedom (with three actuated joints) Phantom Omni models demonstrate the effectiveness of the online neighbor selection approach and compare it to previous work which does not actively select neighbor candidates.