Minimum phase uncertain nonlinear systems in output feedback form are considered: they are subject to disturbances and/or uncertainties and are required to track reference signals. Both disturbances and references are generated by an exosystem whose model and order are uncertain. Assuming that the regulator problem has a solution, but with no a priori assumption on the required control input (e. g. no 'immersion' assumption), a global robust error feedback regulator is explicitly computed which includes an adaptive internal model and a nonlinear stabilizing control, on the basis of known bounding functions for the uncertain system nonlinearities. The regulation error tends to a residual set which decreases as the reference input modeling error due to the internal model choice decreases. If the adaptive internal model can generate the required unknown control input, global asymptotic regulation is achieved.