Reinforcement learning is a powerful framework for the training and evaluation of learning agents. My main focus in this field relates to situations with sparse rewards, in which classical RL algorithms are ineffective.
State representation Learning
Learning proper state representations is considered fundamental for the agents to learn in a robust and efficient way.
Evolutionary algorithms can help to overcome the limits that are present in RL. Moreover, divergent search algorithms like Novelty Search can work as good exploration strategies in situations in which little to no reward is present.