On adaptive inverse dynamics control of rigid robots

The authors derive an adaptive inverse dynamics control law for rigid robots that overcomes the most restrictive difficulty of such schemes to date, namely, the requirement that the inverse of the estimated inertia matrix remain bounded. The control scheme still requires the joint accelerations for its implementation. In practice, one would implement this scheme by estimating the acceleration from the measured velocity. Good performance could still be expected without direct acceleration measurement, assuming sufficient encoder resolution and sufficiently fast sampling. Another practical advantage of the authors' scheme is that since the 'nominal' inverse dynamics or inner loop portion of their control law is fixed, it is more amenable to the development of a dedicated interface and can be 'burned' onto a chip to reduce online computation requirements. >