Stable Sampled-data Adaptive Control of Robot Arms Using Neural Networks
Journal of Intelligent and Robotic Systems
| Hi-index | 0.00 |
The rapid development and formalization of adaptive signalprocessing algorithms loosely inspired by biological models can bepotentially harnessed for use in flexible new learning controlalgorithms for nonlinear dynamic systems. However, if suchcontroller designs are to be viable in practice, theirstability must be guaranteed and their performancequantified. In this paper, the stable adaptive tracking controldesigns employing "neural" networks, initially presented in Sannerand Slotine (1992), are extended to classes of multivariablemechanical systems, including robot manipulators, and bounds aredeveloped for the magnitude of the asymptotic tracking errors andthe rate of convergence to these bounds. This new algorithm permitssimultaneous learning and control, without recourse to an initialidentification stage, and is distinguished from previous stableadaptive robotic controllers, e.g. (Slotine and Li 1987), by therelative lack of structure assumed in the design of the controllaw. The required control is simply considered to contain unknownfunctions of the measured state variables, and adaptive "neural"networks are used to stably determine, in real time, the entirerequired functional dependence. While computationally more complexthan explicitly model-based techniques, the methods developed inthis paper may be effectively applied to the control of manyphysical systems for which the state dependence of the dynamics isreasonably well understood, but the exact functional form of thisdependence, or part thereof, is not, such as underwater roboticvehicles and high performance aircraft.