Extraction automatique d'indices géométriques pour la préhension d'outils en ergonomie virtuelle

Abstract

Advancements in digital technology are reshaping society by offering products and services designed to support and assist people in their daily and professional tasks. Replicating human behaviour is therefore a crucial challenge in robotics and virtual ergonomics. DELMIA is a brand from Dassault Systèmes that is specialized in industrial process simulation. The module enables work tasks to be modelled in 3D simulated manufacturing environments in order to analyze their ergonomics, with the virtual dummy manipulated manually by an expert user. In order to democratize access to virtual ergonomics, Dassault Systèmes has introduced a program that automatically positions the dummy in the virtual environment using a new positioning engine called “Smart Posturing Engine” (SPE). The automatic placement of the dummy’s hands on tools is one of the challenges of this project. The general objective of this thesis is to propose a method for automatically extracting grasping cues, to serve as a guide for tool grasping, based on the tools’ 3D geometric models. This method relies on the natural affordance of the tools that are generally available in a manufacturing environment. The empirical method presented in this study therefore focuses on standard one-handed tools. The method assumes that the tool’s family (mallets, pliers, etc.) is initially known, which makes it possible to assume the affordance of the geometry to be analyzed. The proposed method consists of several steps. First, a section scan is performed on the 3D geometry of the tool. The properties of each section are then extracted to reconstruct a simplified study model. Based on the variation of the properties, the tool is successively segmented into zones, segments and regions. Grasping cues are extracted from the identified regions, including the head of the tool, which provides a task-specific working direction, and the handle or trigger, as the case may be. These grasping cues are finally fed to the SPE to generate task-specific manners of grasping. The proposed solution was tested on fifty one-handed tools belonging to different tool families, such as mallets, screwdrivers, pliers, straight-handle screwdrivers and screw guns. The 3D tool models were retrieved from Dassault Systèmes’s online “Part Supply” site. The proposed method should be easily transferable to other tool families.

Date	31 Jul 2019
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Louis Rivest (Supervisor) & Rachid Aissaoui (Co-supervisor)