Constrained skill discovery paper deploys unsupervised locomotion on ANYmal

Key points

• The paper proposes constrained unsupervised skill discovery for quadruped locomotion using a latent representation over states.
• The authors report deployment of learned policies on a real ANYmal quadruped robot for zero-shot Cartesian-state reaching.
• It matters for ANYmal because it is a direct external research deployment on physical ANYmal hardware.