To improve the task planning module in intelligent service robots, an extension of the action language C+ is proposed and implemented by introducing composite actions as a sequential executions of other actions. The soundness and completeness of the extension is proved by relating the action description in the extended C+ to its corresponding transition system. In the domain of robotic task planning, a composite action can be treated as a “high-level” abstraction of the robot’s physical functions. Such an extension leads to a more intuitive and flexible representation of the robot’s task planning system, and the knowledge of composite actions can be added to the domain incrementally. The experimental results show that for large domains, the extension leads to a great improvement of the solving efficiency.