An operator-aiding manipulator control system

Postural control of mainpulators is the generation of multidimensional time-trajectory data regarding spatial relationships between adjacent pairs of links. Since reduction of the control information dimension facilities manipulator maneuverability, it is proposed that the information in grasping-end position trajectories is mapped into its orientation trajectories. Consequently, during real-time control, the operator needs to provide position information only. Here, the position information has been designated by a multi-dimensional vector. These vectors comprise a pattern space. The orientation information has been designated by a finite set of prerecorded trajectories. These represent particular machine states. A one-to-one correspondence has been established between the categories of the position vectors and the orientation states. A controller model is organized in which the classification of an operator-supplied position vector effects an orientation state transfer. The controller is synthesized by defining a pattern space, a discriminant function, and by developing a training/testing algorithm. The system was tested by programming a 3-dimensional computer-controlled manipulator, which after training required 2-dimensional control input.