where the C, are matrices of Coriolis and centrifugal coefficients. If T is used to denote the torque due to the minimum norm acceleration, then T can be obtained by substituting the first term of (6), i.e., the pseudoinverse solution, into (7), which results in where T is the vector of joint torques, H is the inertia matrix, C is a vector of torques due to Coriolis and centrifugal effects, and g is the gravity vector. The elements of the vector c can be written in quadratic form, so that It has been shown [8] that one can sacrifice the minimum norm acceleration in order to locally minimize the norm of the torque by using a solution of the form which uses a homogeneous acceleration term in a manner analogous to that of (4). The value of this minimum torque can beThe dynamic equations of a manipulator can be written in closed form aswhich has been simplified by taking advantage of the fact that the projection operator is Hermetian and idempotent [14].In all of the above techniques, the specified end effector trajectory is the implicit primary criterion. Unfortunately, the specification of an arbitrary homogeneous joint velocity may result in unrealistic demands on manipulator performance. These difficulties were first illustrated in [11] where the dynamic performance of a redundant manipulator showed significant end effector tracking errors when a secondary criterion was imposed. A more dramatic difficulty with using homogeneous solutions is the instability illustrated in [8] when redundancy is resolved at the acceleration level to instantaneously minimize joint torque. In this case, the joint acceleration is related to the end effector acceleration by differentiating (1) to obtain where once again the general solution is expressed in the form where the subscript 2 refers to the secondary criterion. The overall solution is then given by substituting (3) into (2) to obtain (2) (1)where + denotes the pseudoinverse, (l -J+ J) is a projection operator onto the null space of J, and ci > is an arbitrary vector in iJ space. The second term in (2) is the homogeneous solution to(1) since it results in no end effector velocity and will be denoted here by iJH' This homogeneous solution is frequently used to optimize some secondary criterion under the constraint of the specified end effector velocity by choosing cP to be the gradient of some function of (J [13]. Alternative formulations for instantaneously optimizing a secondary criterion by augmenting the Jacobian matrix have also