Simulation of human activities in smart spaces is becoming increasingly important due to the growing demands on datasets that captures user technology interactions. Such datasets are critical to the success of human-centric research that relies on understanding user actions, activities and behavior in a given situation and space. But most existing activity models have been designed for the purpose of recognition and are not specifically optimized for visual simulation (playback) or for dataset generation. To address this limitation we propose an activity playback model and associated algorithms to facilitate activity simulation with high degree of realism and with inherent variability that mimics the human nature of performing the same activity variably. We present the activity playback model, algorithms, and a validation study to assess the realism of our approach. © Springer International Publishing Switzerland 2015.