The transient problem in self-tuning control is analysed in a non-stochasti
c framework based on model unfalsification and solutions are proposed. It i
s assumed that the disturbance bounds or norm bounds on the unmodelled dyna
mics are not known a priori. Depending on the transient requirements the pr
oblem is grouped into four classes. Simulations show the practical potentia
l of the tuning methods proposed.