It is argued that the tendency of nature to minimize energy may be used as
a unifying basis for all robust estimators. Robustness is defined and discu
ssed based on mechanical rather than empirical and abstract tools. This mec
hanical view of robustness is then extended to design new and useful robust
data editors that suppress the outlying content of the contaminated observ
ations. These editors are applied to edit samples of sea-surface heights, g
ravity observations and reduced global positioning system baselines.