The NASA Terrestrial Planet Finder Interferometer (TPF-I) and ESA Darwin missions are designed to directly detect mid-infrared photons from earth-like planets around nearby stars. The technique of nulling interferometry is used to suppress the light from the parent star, typically 107 times brighter than the planet, with an angular offset of 10-100 mas. There are two classes of noise: photon shot noise from the stellar leakage, local- and exo-zodiacal dust, and instability noise from variations in the instrument response. Shot noise requires that the instrument null depth is at least 10-5. The instability noise requires a null depth of ~10-6, corresponding to control of the optical paths at ~1 nm rms, and control of relative intensities at ~0.2%; it is these requirements that currently drive the design of the instrument. This paper describes a new technique that effectively removes the effect of instability noise. This relaxes the nulling requirement by a factor ~10 and makes planet detection robust to instrument variations. At the same time, the integration period needed to detect and characterize planets is reduced and the angular resolution of the array is significantly improved. Analysis and simulations are presented, and implications for the array architecture are discussed.