Evaluation de l’autonomie des personnes à l’aide de dispositifs de type intelligence ambiante : création et validation d’un nouveau test écologique et modélisation de son analyse, une approche ergonomique du processus d’automatisation.

The diagnosis of the dysexecutive syndrome is commonly based on the measurement of the patient's performance in performing tasks in the performance of tests, derived from laboratory tasks aimed at separately evaluating the different executive functions of the latter. Despite their clear therapeutic value, these tests have significant limitations. These performances are often poor predictors of the patient's ability to perform tasks that mobilize these same daily functions to lead an independent life. A few tests tackle the systematic observation of the patient's behavior in everyday life to solve the issue of ecological validity, following an approach similar to that developed in the ergonomics of the activity. One of these, the Cooking Task test, concerns the preparation of a cake from the follow-up of a cooking recipe (Chevignard et al., 2008). A typology of cognitive errors, whose frequency can be considered an indicator of the severity of the dysexecutive syndrome, serves to analyse the results. One of these, the Cooking Task test, concerns the preparation of a cake from the follow-up of a cooking recipe (Chevignard et al., 2008). A typology of cognitive errors whose frequency can be considered an indicator of the severity of the dysexecutive syndrome.This doctoral thesis presents state-of-the-art knowledge relating to executive functions and diagnostic tests for the dysexecutive syndrome. Although the most recently developed methods tend towards greater ecological validity, an analysis of their limits led us to propose an alternative test based on constructing a complex building using Lego® bricks, called the "Arc de Triomphe Construction Test" (ATCT). We carried out a systematic observation of behaviour and visual information taking in 63 patients – 43 of whom had head trauma and 20 with post-radiation leukopathy – performing this test, and we compared their performance with the Cooking Task test. The results show a higher validity of the ATCT compared to the existing test. We discuss these results.Despite these encouraging results, a limitation to the widespread test adoption comes from the cost of systematic observation because this method is time-consuming when done manually. A complementary study proposes to analyse the coding activity of a video recording of the behaviour of a patient by eight participants, ergonomists trained in the use of coding software. This analysis produced a model of this activity. We discuss the prospects of developing tools that exploit Artificial Intelligence for image recognition to automate, in part or in full, the coding of behavioural data.