Learning to solve MuJoCo locomotion tasks from a single state: Hopper