Nonconvex Noise-Tolerant Neural Model for Repetitive Motion of Omnidirectional Mobile Manipulators

Dear Editor, Quadratic programming problems (QPs) receive a lot of attention in various fields science computing and engineering applications, such as manipulator control [1]. Recursive neural network (RNN) is considered to be powerful QPs solver due its parallel processing capability feasibility hardware implementation [2]. In particular, large number RNN models, gradient network, are proposed alternatives for online solving [3]. However, it worth noting that most the above networks essentially designed static with time-invariant parameters. These algorithms cannot solve time-varying (TV) because they adapt change parameters, kinematic redundant arms [4]. Zeroing (ZNN) specially real-time solution problems. It uses time derivative (TD) parameters zero-finding problem [5]. The Taylor-type discrete-time ZNN (DTZNN) model [6], which outperforms other models inherently used address QPs, Newton iterations. Although DTZNN makes full use TD information solved, still does not explicitly consider influence noise. nonlinear system, there system errors or external disturbances implementation, can regarded noise [7]. Different constructed by choosing different error functions (EFs) utilizing activation (AFs) existing but design process roughly similar. AF should monotone increasing odd function. Therefore, ZNN-based drawn relaxing convex constraint TVQPs equality inequality constraints (EAICs) presence noises.