Neural-network-based direct adaptive controller design for nonlinear robotic manipulators

This paper presents an all-pass H_∞ controller design with and without the enhancement from the neural-network-based direct adaptive scheme for nonlinear robotic manipulators with particular examples from class 1 and class 4 robot manipulators. The all-pass H_∞ gains for the robust adaptive controllers need not be proper, and thus may resemble the widely-used proportional and derivative controllers. This all-pass H_∞ gain-designing approach is purely based on the decoupled SISO transfer function analysis of the robotic motion. To make the robotic manipulator more versatile and trainable, a neural-network-based direct adaptive scheme is introduced in addition to the all-pass filters. The neural-network-based adaptive feedback system is proved to be exponentially stable. The learning rates for the neural-network architecture are determined analytically by employing Lyapunov stability analysis of the feedback nonlinear systems