Communication-efficient multi-robot exploration using coverage-biased distributed Q-learning