Using Neural Networks to Avoid Robot Singularity

The seminal study presented by Yoshikawa exploits the pseudoinverse‐type formulation to avoid the singularity of redundant manipulators by maximizing the manipulability measure. This chapter models the manipulability optimization scheme of redundant manipulators and proposes dynamic neural network to remedy the weaknesses of existing quadratic programming formulation based manipulability‐maximizing scheme. It presents definitions on manipulator kinematics and manipulability for problem formulation. The chapter reformulates a redundancy resolution problem with optimal manipulability considered into a constrained quadratic programming problem. It develops neural dynamics for real‐time solution of the optimal manipulability redundancy resolution problem by converting the problem into the solution of a nonlinear equation set and then establishing neural dynamics for solving the nonlinear equation set. The chapter discusses computer simulations that are conducted on a PUMA 560 manipulator to demonstrate the effectiveness of the proposed manipulability optimization scheme as well as its dynamical neural network solver.