This paper is a brief report on the recent improvement of a novel approach invented for the control of approximately and partially known multivariable, non-linear, strongly coupled mechanical systems under dynamic interaction with an unmodeled environment. Like traditional Soft Computing (SC) does this method uses simple uniform structures and standard procedures. In contrast to the traditions, instead of Kolmogorov's approximation theorem these special structures originate from the mathematical framework of the Lagrangian Mechanics. This results in a considerable reduction of modeling complexity since the number of the tunable parameters is given ab ovo by knowing only the degree of freedom of the system to be controlled. This fact simplifies parameter tuning or "learning". Further advantage is that the great majority of "scaling" problems can be avoided in the case of the structures here used. The operation of the method is illustrated by simulation results developed for 3 degree of freedom a SCARA robot.