Towards Online Testing Under Uncertainty Using Model-Based Reinforcement Learning

Modern software operates in complex ecosystems and is exposed to multiple sources of uncertainty that emerge in different phases of the development lifecycle, such as early requirement analysis or late testing and in-field monitoring. This paper envisions a novel methodology to deal with uncertainty in online model-based testing. We make use of model-based reinforcement learning to gather runtime evidence, spot and quantify existing uncertainties of the system under test. Preliminary experiments show that our novel testing approach has the potential of overcoming the major weaknesses of existing online testing techniques tailored to uncertainty quantification.