SLR: Learning Quadruped Locomotion without Privileged Information