We argue that the development of robots that can interact effectively with people requires a special focus on building systems that can perceive and comprehend intentions in other agents. Such a capability is a prerequisite for all pro-social behaviour and in particular underpins the ability to engage in instrumental helping and mutual collaboration. We explore the prospective and intentional nature of action, highlighting the importance of joint action, shared goals, shared intentions, and joint attention in facilitating social interaction between two or more cognitive agents. We discuss the link between reading intentions and theory of mind, noting the role played by internal simulation, especially when inferring higher-level actionfocussed intentions. Finally, we highlight that pro-social behaviour in humans is the result of a developmental process and we note the implications of this for the challenge of creating cognitive robots that can read intentions.