Robust optimal neural control of robots

In this work possibility of improvement of algorithms of neurocontrol of anthropomorphic manipulators is researched. Problems solved include: synthesis of neural stabilizing and optimal control algorithms with improved performance and robustness; and simulation of achieved results. New algorithms of optimal in time neural control of manipulators are based on global decomposition of dynamic model, considering the constraints on accelerations of the robot joints. The proposed combination of traditional nonlinear control algorithms and neural algorithms of dynamic model approximation ensures improved control quality (required transient parameters). The proposed method makes possible to compensate dynamics approximation errors, thus providing robust control. Obtained robustness estimates for developed neurocontrol algorithms establish relation between transients quality and parameter disturbances, caused by inaccurate approximation. The earlier obtained results (1996) are generalized and supplemented with new details