Open Access Database of Industry 4.0 Tasks for the Development of AI-Based Classifier

Robots and humans coworkers are sharing more and more portions of the smart manufacturing globally, meeting the need for high flexibility and rapid changes in the production layout. To be fully effective, however, such transition from classic robotics to the so-called collaborative robotics has to address several open problems, mostly related with safety and task optimization. Promising answers are coming from the motion capture technology, where wearable and optoelectronic sensing devices are deployed to gather human centric data to provide the robots with some form of awareness respect with the human activity and position. Tracking the hand of the operator, in particular, offers many advantages as we use our hands to explore and interact with the surroundings and to communicate. This has been highlighted by the several works focusing on gesture hand configuration recognition. This work present HANDMI4, a new open access database of hand motion tracking data, which includes a wide range of static hand grasp configurations and some classic dynamic industry tasks. Such database was generated using two of the most mature technologies for motion capture: IMU-based data glove and camera-based triangulation. To test the capability of such dataset to foster AI-based task classifier, a set of machine learning techniques were implemented and tested. In particular, KNN weighted reached 94,4% and 100% of task classification accuracy for the data glove and the camera system, respectively. With this open access database we aim to boost the research around task classification through motion capture technology to enable the next revolution in smart manufacturing.