Real-time fine motion control of robot manipulators with unknown dynamics (Q2769555)

I cannot resist an impression that this paper is nothing more but another instance of ``new skin for the old ceremony''. Dealing with tracking control in the jointspace for rigid robotic manipulators, the paper re-states two well-known results in terms of neural networks. The first result is almost the same as the adaptive computed torque algorithm (``almost'' means that the dependence of control on the desirable acceleration has been skipped). The second result is almost the same as the Slotine-Li adaptive algorithm control (``almost'' means the same as above plus additionally using the current instead of the reference joint velocity). Proofs of convergence seem to have overlooked the fact that the error dynamics depend on time. In both cases the well-known parameter update laws have been obtained and advocated to be realized by a neural network. Simulation examples illustrate the theory.