Beyond thinking in common categories: Predicting obstacle vulnerability using large random codebooks | IEEE Conference Publication | IEEE Xplore