Complex software systems require frequent testing to ensure their functionality and to discover flaws in their implementation early after they have been introduced. However, executing all tests after each change is often not feasible, especially when physical test agents with limited resources, e.g. industrial robots, are involved. In this talk, we discuss the usage of machine learning and constraint programming for efficient test management. We see how test cases are prioritized with reinforcement learning, and how the test plan is scheduled, such that tests are executed on different physical systems over time.