This paper focuses on the well-known problem in behavioral robotics - "what to do next". The problem addressed here lies in the selection of one activity to be executed from multiple regulative, homeostatic and developmental processes running onboard a reconfigurable multi-robot organism. We consider adaptive hardware and software frameworks and argue the non-triviality of action selection for evolutionary robotics. The paper overviews several deliberative, evolutionary and bio-inspired approaches for such an adaptive action selection mechanism.