This chapter presents the learning capabilities that have been introduced to the ASTRA agent programming language. Using learning templates, agents can acquire new plans at runtime using different Machine Learning algorithms while exerting control over the learning processes. Algorithms specified in the learning templates operate over the agents belief set to create new plans with different contexts, according to the state of the environment at the time of learning, as modeled by the agents belief set. Agents leverage control over the learning processes using protected belief predicates and internal actions. The learning templates allow different algorithms to be used: one agent can use multiple algorithms to learn different plans. Agents are also able to share plans learned at runtime between them, during the learning process, or after it has been completed.