A wide range of reinforcement learning (RL) problems -- including robustness, transfer learning, unsupervised RL, and emergent complexity -- require specifying a distribution of tasks or environments in which a policy will be trained.

Deep reinforcement learning (RL) agents may successfully generalize to new settings if trained on an appropriately diverse set of environment and task configurations.

The first theorem is the main result that allows us to analyze dual curriculum games. We provide a formal proof below: Proof. We get the desired result by combining these two arguments. Repeating the symmetric argument shows the desired property for the second teacher. Following this main theorem, we can apply it to two of our methods.

Deep reinforcement learning (RL) agents may successfully generalize to new settings if trained on an appropriately diverse set of environment and task configurations.

We call the problem of choosing how to generate these environments UED.