This paper investigates, for two distinct classes of flexible joint ro
bot arms, the problem of achieving exact output tracking over a finite
time interval by means of a learning strategy. First, the problem of
initial conditions compatibility with the trajectory to be tracked is
discussed. Next, the author shows that after high-gain feedback linear
ization, the control needed to track a family of link trajectories tha
t satisfy some constraints on the initial conditions can be learned by
means of a P-type algorithm.