In an agent system that needs to operate in a real world, the problem of maintaining a consistent world model in the face of unreliable, incomplete and inconsistent sensory data should be solved. In this paper, we present an approach that addresses this problem by applying an argumentation-based scene interpretation framework for accurately modelling and representing the observations and beliefs of an agent. Our approach is based on temporal and probabilistic defeasible logic programming for reasoning. The performance of our approach is evaluated on simulation experiments in the Stage Robot Simulator. We also show that our approach is applicable to real world scenarios with an autonomous Pioneer 3-AT robot.