We propose priori knowledge-free algorithms for distributionally robust average-reward estimation, with both finite-sample theoretical guarantees and numerical validation.