Définition de la préhension en ergonomie virtuelle : vers une démarche automatique basée sur la similarité géométrique des objets

Abstract

Computer simulation, a vast field, is being used more and more in the development of products and manufacturing processes. Thus, it is now possible to simulate a work environment to, among other things, evaluate ergonomics. The work of positioning the virtual manikin within a digital mock-up remains however very interactive and not very intuitive. That's why Dassault Systèmes has launched a project to develop a new positioning engine for the manikin. Among the objectives of this project is the automation of the positioning of the manikin, which requires the automation of the definition of the grasping of objects by the virtual manikin. The general objective pursued in this thesis is to evaluate an approach based on geometric similarity and the transposition of the grasp to similar objects, in order to provide the new positioning motor of the manikin a plausible grasp even for an unknown tool of the manikin. This approach involves several steps. When an unknown object is presented to the manikin in the virtual environment, the system looks in a database of typed objects for an object geometrically similar to the new object. The system will then transpose the grasp information specific to this typed object to the new object, to transmit this grasp information to the positioning engine of the manikin. In order to evaluate this solution, a database was first created. First of all, it was necessary to choose a list of test objects, in this case, the usual industry tools, such as a hammer, a drill, a screwdriver, a combination wrench, and so on. For each of the selected tools, one or more actions are considered. Indeed, there may be, on the one hand, several ways to use the same tool. It may also be necessary, on the other hand, to define 'generic' actions, such as 'grab to move' the tool. In a second step, we checked the impact of anthropometry on the grasp. Since we want to record, for each typed object, a tool-action pair, it was necessary to determine whether a plausible grasp for a 50th percentile could also be plausible for a woman at the 5th percentile or a 95th percentile. If an average reference grasp had proved unusable for distant anthropometries, it would have been necessary to include tool / action / anthropometry triplets in the database. The various tests carried out show that the impact of anthropometry on grasp is limited and that it is appropriate to include only the grasp of a man at the 50th percentile within the database. Finally, it was necessary to evaluate the possibility of transferring the grasp of a tool from the database to an unknown but geometrically similar tool. The first step is to find in the database a tool known geometrically similar to the new tool. Dassault Systèmes offers its own geometric search engine, and we wanted to test this solution. We have seen encouraging results, however, some tests indicate that there is still work to be done. In the hypothesis that the search for geometrically similar tools is going well, we wanted to evaluate the ability to transfer the grasp information of the typed tool to the geometrically similar tool. The method examined consists of predicting geometric elements (plane of symmetry, etc.) on the typed objects, and trying to find equivalent geometric elements on the new geometrically similar object. These geometric elements are then used as resetting constraints between the two tools. Once the registration is completed, it is possible to transfer the grasp of the typed object to the new object.

Date	5 Jun 2018
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Louis Rivest (Supervisor)