Active learning approach co-navigates theory and experiment via hybrid of reinforcement learning and structured Gaussian Processes