AI4Assembly: AI-supported, collaborative assembly of complex, mechatronic modules

Problem definition

Human-robot collaboration (HRC) usually requires complex and costly teach-in process performed by professionals, in order to add or change some assembly sequences or robot trajectories. In addition, classical methods for object detection and localization are inefficient and limited. Furthermore, a safe and collision-free HRC must be ensured.

Goal

Safe and intelligent HRC assembly process, through AI-supported teach-in processes and object localization.

Solution expertise

AI is used as a "service layer" to simplify configuration processes. Learning from human demonstration is enabled by deriving assembly sequences via trained classifier and by intelligently capturing specific robot motions and trajectories.

Classical methods for object localization are complementary improved with AI methods, so that a fast adaptation to new objects is possible with little training data and an efficient and exact localization is guaranteed as a 6D Pose.

AI-based systems also significantly improve collision avoidance in the MRK scenario, through a volumetric representation method for virtualizing the workspace using GPU voxels, and a prediction of human movements.

Test bed

Real setup as hands-on is the basis for different industrial applications.