Introduction: Autism Spectrum Disorder (ASD) is a by-birth neurodevelopmental disorder difficult to diagnose owing to the lack of clinical objective and quantitative measures. Classical diagnostic processes are time-consuming and require many specialists’ collaborative efforts to be properly accomplished. Most recent research has been conducted on automated ASD detection using advanced technologies. The proposed model automates ASD detection and provides a new quantitative method to assess ASD.Methods: The theoretical framework of our study assumes that motor abnormalities can be a potential hallmark of ASD, and Machine Learning may represent the method of choice to analyse them. In this study, a variational autoencoder, a particular type of Artificial Neural Network, is used to improve ASD detection by analysing the latent distribution description of motion features detected by a tablet-based psychometric scale.Results: The proposed ASD detection model revealed that the motion features of children with autism consistently differ from those of children with typical development.Discussion: Our results suggested that it could be possible to identify potential motion hallmarks typical for autism and support clinicians in their diagnostic process. Potentially, these measures could be used as additional indicators of disorder or suspected diagnosis.

In this paper we compare systematically the most promising neuroevolutionary methods and two new original methods on the double-pole balancing problem with respect to: the ability to discover solutions that are robust to variations of the environment, the speed with which such solutions are found, and the ability to scale-up to more complex versions of the problem. The results indicate that the two original methods introduced in this paper and the Exponential Natural Evolutionary Strategy method largely outperform the other methods with respect to all considered criteria. The results collected in different experimental conditions also reveal the importance of regulating the selective pressure and the importance of exposing evolving agents to variable environmental conditions. The data collected and the results of the comparisons are used to identify the most effective methods and the most promising research directions.

We demonstrate how the evolutionary training of embodied agents can be extended with a curriculum learning algorithm that automatically selects the environmental conditions in which the evolving agents are evaluated. The environmental conditions are selected to adjust the level of difficulty to the ability level of the current evolving agents, and to challenge the weaknesses of the evolving agents. The method does not require domain knowledge and does not introduce additional hyperparameters. The results collected on two benchmark problems, that require to solve a task in significantly varying environmental conditions, demonstrate that the method proposed outperforms conventional learning methods and generates solutions which are robust to variations and able to cope with different environmental conditions.

In this paper we analyze the role of environmental variations in the evolution of artificial agents situated in an external environment. By comparing the results of experiments carried out in different conditions we demonstrate how environmental variations promote the evolution of better solutions. The beneficial effect is maximized at intermediate rates of variations, i.e. when the dynamics of the environment displays a sufficient level of stability and variability. The advantage of intermediate rates of variation is due to the fact that the occurrence of new environmental conditions favors the retention of phenotypic variations that can eventually provide an advantage also in different environmental conditions. Moreover, a relative stability of the new environmental conditions constitutes a prerequisite for the generation and retention of such variations.