Robot assistants act in dynamic environments and have to cope with changing situations: they must be able to sense the world and to control their activities in real-time. They must also be able to execute complex, multi-step plans, to confirm that plan steps have had their intended effects, and to deal with problems and opportunities as they arise. This paper presents a modular control architecture for task planning and execution, that controls concurrent, reactive, and deliberative activities. The robustness against changing environments is demonstrated by Care-O-bot II showing fetch-and-carry duties. Experimental results are reported.