Conflict prediction in human-machine systems
Prédiction des conflits dans des systèmes homme-machine
Résumé
The work is part of the research dedicated to human-machine misunderstandings and automation surprises, and conflicts between the human and the machine that may arise from such situations. Conflicts in a human-machine system stem from the fact that both agents can decide and perform actions that may be inconsistent, either because the expected plan for the human or the machine is not followed anymore, or because the human has a wrong situation awareness, or both. In order to prevent future mission degradation possible conflicts have to be detected early in the design phase so as to allow redesign, interface adaptation or specific training to be considered. A Petri-net based conflict pattern dedicated to conflict identification in human-machine system models is proposed: the pattern a priori identifies state transitions that are likely to bring about conflicting situations. An experiment with pilots was conducted to test the soundness of this formal approach: the results show that these transitions indeed generate human-machine conflicts that may remain unnoticed or unsolved by the participants. Starting from this model we further propose a possibilistic development dedicated to real time detection and identification of human-machine conflicts. Critical situations may arise also because of human error. While it is possible to design systems to be less vulnerable to errors, the systems also require the ability to gracefully recover from errors when they do happen, reversing their effects. In the frame of the reversibility assessment we define a reversibility scale and we develop reversibility check in order to assess the degree of reversibility of the human operator'actions.
Prédiction des conflits dans le systèmes homme-machine. Le travail fait partie de la recherche consacrée à des problèmes d'interaction homme-machine et aux conflits entre l'homme et la machine qui pourrait découler de ces situations.
Loading...