In this paper, the problem of signal detection in severe and/or changing noise environments, often encountered in underwater acoustics and communications, is considered. The detectors operate in sequential, i.e., variable sample size, mode at near optimum levels for a particular noise environment and are robust by maintaining high efficiency in other than the nominal noise environments by adapting their optimum nonlinearity using anm‐interval polynomial approximation (MIPA) of it. Furthermore, the assumption of independent samples is relaxed, allowing higher transmission rates for lower error rates. The proposed sequential detectors work well for coherent and noncoherent detection, are asymptotically optimum and, for small signal‐to‐noise ratios (SNR), increase their transmission rate up to four times as compared to their fixed‐sample size counterparts. The performance of the detectors is evaluated in typical underwater noise fields. It is demonstrated that their efficiency with respect to sequential MIPA detectors with independent sampling is improved, in some cases up to the order of the dependence of the samplesQ. It is emphasized that the MIPA detectors are easy to implement. The estimation and updating of the detector parameters may be accomplished using parallel processors operating in a recursive mode without disturbing the decision process.