The problem of moving a rigid robot arm along a finite sequence of equ
ilibrium points, with the last point coincident with the first one, is
investigated. Such a sequence, referred to as a cycle, is to be repea
ted over and over in time, and a controller is sought which improves s
ystem performance by using positioning errors. Differently from learni
ng control, no system initialization is required at the end of trial.
After high gain feedback linearization of the robot dynamics, it is sh
own that linear, robust, finite dimensional algorithms can be set up t
o accomplish this task for unconstrained robots and robots subject to
smooth bilateral constraints for which hybrid force control is of inte
rest. An experiment on a two-link robot arm illustrates algorithm appl
icability.