
Exploiting the contextual structure of objects and the compositional nature of temporal logic task specifications, this approach generates auxiliary tasks that share similar underlying exploration requirements with a given target task, enabling efficient off-policy learning.


coremsg

leveraging-contextual-object-embeddings-and-temporal-logic-to-generate-useful-auxiliary-tasks-for-reinforcement-learning


Leveraging Contextual Object Embeddings and Temporal Logic to Generate Useful Auxiliary Tasks for Reinforcement Learning
