Modeling the behavior of persons with mild cognitive impairment or Alzheimer’s for intelligent environment simulation

Abstract

Intelligent environments may improve the independence and quality of life of persons with mild cognitive impairment (MCI) or Alzheimer’s disease (AD) through their ability to automatically provide assistance or guidance. In order to deploy these systems in this category of the population, it is necessary to be able to carry out validation and experiments to improve efficiency, safety, user experience and reduce installation costs. Unfortunately, this type of experiment can be difficult to perform because of the difficulty in recruiting candidates and accessing adequate intelligent environments. These problems could be partially offset with simulators. These tools can be used to simulate the behavior of an intelligent environment and its occupants in order to generate data, or to observe and evaluate their behavior. However, to design systems for populations suffering from MCI or AD, it is necessary that the simulator be able to emulate the behavior of these persons. In this paper, two approaches to simulate and generate sequences of actions containing errors usually committed by persons with this type of disease are proposed. Those approaches aim to be simple to use and both are based on the use of behavior trees. The first one consists in adding nodes to a behavior tree to simulate errors with their specific probabilities. The second approach consists in defining an interval to bind the number of errors that can be inserted through the error injection algorithm. We also present the results of the experiments carried out to evaluate these approaches. For the first experiment, several simulations were conducted and were recorded in videos. These videos were analyzed by specialists in cognitive disorders who diagnosed the avatar of these videos. The second experiment aimed at comparing the two approaches together. To do so, several action sequences were generated. The results show that our model is able to generate healthy, MCI and Alzheimer’s behaviors. The results also show that the second approach facilitates the generation of a desired number of errors.