The predecessor projects of this proposal were highly successful and generated numerous publications at flagship AI journals and conferences. Furthermore, they led to winning several awards for software systems at international conferences. The latest project even won the prestigious ICAPS Best Paper award. In this continuation project, we will keep the original project focus and build on our past results to tackle new research challenges. Connecting the fields of model-based reasoning and data-driven learning has recently been identified as one of the key research goals in artificial intelligence. Our project will contribute to this endeavor, focusing on the area of automated planning. We will learn heuristic functions that guarantee optimal solutions and planning algorithms that dynamically adapt to the given task. Automated planning is the task of finding a sequence of actions that lets an agent achieve their goals. We focus on the classical planning setting, where the agent's world is fully observable and all action outcomes are deterministic. Today's classical planning systems are powerful enough to solve challenging, real-world planning tasks. The strongest method for solving classical planning tasks is state-space search, where we search for a solution path in a directed graph of world states. This method strongly relies on heuristics, that is, functions that estimate how far a state is away from the nearest goal. Planning researchers have manually devised a large collection of different heuristics and state-space search algorithms. These algorithms are able to solve challenging planning tasks, but they usually require a lot of work to be devised, understood and implemented. More importantly, the design spaces for heuristics and search algorithms are huge and it is infeasible to manually explore them in any meaningful way. In this project, we will use machine learning to learn heuristics and search algorithms automatically.