Home-service robots are expected to perform a wide range of tasks commonly encountered in household environment. For autonomous operations robots should possess intelligence to plan the actions to carry out these tasks from the beginning, or they should at least have the ability to learn how to plan for more tasks during their operation. Since it is impossible to predict all tasks in advance and write programs for robots to perform the tasks, it is best to endow robots with a learning capability. We use a case-based reasoning approach to home-service-robot learning because of the richness and diversity of information needed fur task planning. Given a new task, a robot finds the closest task among the tasks it blows how to plan for, and it modifies the plan to adapt to the new task. To expedite the reusability and performance of the task planning, we design a RTML (Robot Task Makeup Language) to represent each task using an Atomic Action Taxonomy  and employ a rule-based reasoning system  to perform a parameter adaptation. After performing a series of processes, which are needed in the robot``s task planning, we can acquire a sequence of atomic actions for a task in the given context.