Learning Control of Robot Manipulators in Task Space

Abstract

Two important properties of industrial tasks performed by robot manipulators, namely, periodicity (i.e., repetitive nature) of the task and the need for the task to be performed by the end-effector, motivated this work. Not being able to utilize the robot manipulator dynamics due to uncertainties complicated the control design. In a seemingly novel departure from the existing works in the literature, the tracking problem is formulated in the task space and the control input torque is aimed to decrease the task space tracking error directly without making use of inverse kinematics at the position level. A repetitive learning controller is designed which "learns" the overall uncertainties in the robot manipulator dynamics. The stability of the closed-loop system and asymptotic end-effector tracking of a periodic desired trajectory are guaranteed via Lyapunov based analysis methods. Experiments performed on an in-house developed robot manipulator are presented to illustrate the performance and viability of the proposed controller.

Towards realizing robotic potential in future intelligent food manufacturing systems

“…Recent state-ofthe-art on current and emergent control strategies for robotic manipualtors are reviewed in [40][41][42][43]. Some intelligent control schemes including artificial neural network, neuro-fuzzy control, expert system, learning control systems for robot manipulator control are also found in literature [44,45].…”

Improved recurrent neural networks for online solution of Moore‐Penrose inverse applied to redundant manipulator kinematic control

“…As shown by Example 2, the proposed model NNN-R (8) and ZNN-R model (6) work well even if the matrix to be inverted is close to rank-deficient (i.e., highly ill-conditioned). While for the GNN-R model (4), it may take a very long time to converge to the theoretical right Moore-Penrose inverse owing to the minimum eigenvalue of AA T being a very small number, and thus the convergence curve of the residual error of the GNN-R model is omitted in Figure 4.…”

“…Numerous applications in mathematics and engineering fields are closely related to online solution of the Moore-Penrose inverse (also termed pseudo inverse), e.g., signal processing [1], robotics [2][3][4][5][6][7][8], and control theory [9,10]. In mathematics, given any matrix A ∈ R m × n , the Moore-Penrose inverse A + is unique, and can be derived from the following four matrix equations [11]:…”

UDE-based task space tracking control of uncertain robot manipulator with input saturation and output constraint

“…At present, there are two main methods to realize the trajectory tracking control of robot manipulators in task space. One method is to use the pseudo-inverse of the Jacobian matrix [17,18,19,20,21,22]. Liu et al [17] proposed an adaptive NN controller for robot manipulators with the optimal number of hidden nodes and less computation.…”