This article presents an improved numerical algorithm for robot inverse kinematics which is based upon the solution of the first-order differential equations arising from the manipulator's velocity Jacobian relations. The use of the Adams-Moulton predictorcorrector scheme leads to a fourth-order trajectory following in the joint space. The implementation of a strict descent feature for the trajectory error at the end-effector level contributes to the robustness of the algorithm near singularities. The execution of this algorithm is about 2.7 times faster than that of Gupta and Kazerounian,I6 with much of the speed up coming from the use of software optimizations. Several issues related to the accuracy, convergence, speed, real-time computation, and portability of this algorithm are discussed.