The authors employ energy-based arguments to establish a robustness and a convergence result for a gradient-descent adaptive law in the presence of both noisy measurements and uncertainty in the initial guess.