The past two decades have seen incredible advances in practical motion planning algorithms. As perception improves and robust robotics hardware becomes a commodity, modern motion planning is enabling new applications with deep societal impact such as transportation, agriculture, and healthcare. A critical next step is to produce motion from high-level specifications. The latter declare what the robot must do, rather than how the task is to be done. This talk will present recent work in developing integrated frameworks that exploit methods from logical reasoning, optimization theory, artificial intelligence and robotics to produce solutions in manipulation scenarios, and will discuss the challenges that lie ahead.