Similar Items: Tackling Uncertainty in Reinforcement Learning: A Dual Variational Inference Approach for Task and State Estimation