Training multiple agents with one demonstration