As we are deploying planning mechanisms in real-world applications, such as the control of autonomous robots, it becomes apparent that the performance of plan-based controllers critically depends on the design and implementation of plan libraries. Despite its importance the investigation of designs of plan libraries and plans has been largely ignored. In this paper we describe parts of a plan library that we are currently developing and applying to the control of a simulated household robot. The salient features of our plans are that they are designed for reliable, flexible, and optimized execution, and are grounded into sensor data and action routines. We provide empirical evidence that design criteria that we are proposing have considerable impact on the performance level of robots.