Carma is a recently developed, high-level quantitative modelling language developed for the design and analysis of collective adaptive systems. In the current Carma language, agents within a system consist of a behaviour, captured as a process, and knowledge, represented as a store of attributes. In this paper we present the first steps to equipping agent specifications with more sophisticated forms of knowledge in terms of goals and targets, and demonstrate how these may be integrated into the modelling and analysis process. We illustrate the ideas with a simple example taken from the domain of swarm robotics.