Spatial understanding and communication are essential skills in human interaction. An adequate understanding of others' spatial perspectives can increase the quality of the interaction, both perceptually and cognitively. In this paper, we take the first step towards understanding children's perspective-taking abilities and their tendency to adapt their perspective to a counterpart while completing a task with a robot. The elements used for studying children's behaviours are the frame of reference and perspective marking, which we evaluated through a task where players needed to compose instructions to guide each other to complete the task. We developed the interaction with an NAO robot and analyzed the children's instructions and their performance throughout the game. Our initial findings demonstrated that children tend to compose their first instruction by following the principle of least collaborative effort. Children significantly changed and adapted their perspective, i.e. frame of reference and perspective marking to the robot, mainly when the robot failed to follow their instructions correctly. Additionally, results show that children tend to create a mental model of their counterparts and the robot changing that frame of reference might affect their performance or the flow of the interaction.