Dynamical Distance Learning for Semi-Supervised and Unsupervised Skill Discovery

DDL is an auxiliary task for an agent to learn distances between states in episodes. This can then be used further to improve the agent’s policy le...