 This paper presents a novel approach to controlling quadruped robots using deep reinforcement learning. The authors developed a learning-based controller which uses a commanded velocity reference to generate a joint impedance reference. This allows the robot to be more efficient with its energy usage and easier to deploy. The authors tested the controller on the Solo 12 quadruped robot and showed that it was able to learn how to walk on different terrains both indoors and outdoors. The results demonstrate that the Solo 12 robot is a suitable platform for research into learning and control due to its ease of transferring and deploying learned controllers. This article was authored by Michel Ariktingi, Pierre Alexandre Lesiat, Thomas Flayors, and others.