In the first, a simulated environment is used to learn a policy that solves the cube-stacking task with synthetic images and proprioception (awareness of the position and movement). Two agents ...