Object Manipulation and Mannequin Driving Based on Multi-Agent Architecture

Abstract

Digital mock-ups, path planners and virtual reality tools are increasingly used within the industrial framework, allowing accessibility and maintainability checks without physical mock-ups. For this, it is necessary to simulate an object moving without collision. Two methods are commonly proposed: automatic path planning approach and direct simulation through robotic or virtual reality tools. In order to use the potential and avoid the drawbacks of these two methods, we settled for a mixed approach using local algorithm abilities and the global view ability of a human operator. To achieve this co-operation, we use a multi-level tree architecture composed by multi-agent “blackboard” systems. The different agents acts to nullify the collisions between the manipulated object and the cluttered environment, attract it towards its target, allow human operator’s actions on the path, and provide the mannequin with a coherent behavior. Thanks to our system we manage to perform an accessibility task for object and mannequin with more than 20 degrees of freedom in a cluttered environment.