From raw signals to human skills level in physical human-robot collaboration for advanced-manufacturing applications
Résumé
Providing individualized assistance to a human when she/he is physically interacting with a robot is a challenge that necessarily entails user profiling. The identification of the human profile in advanced-manufacturing is only partially addressed in the literature, through either intrusive, or not fully transparent approaches. As on-the-job training has a negative impact on operators’ working conditions, we specifically focus on their skills, and show that they can be observed in a non-intrusive way, through a data-driven approach to extract knowledge from the internal data of the robot. To this end, we have defined useful characteristics derived from raw data, called in this paper Key Skill Indicators (KSI), and have devised a user’s skills model based on expert knowledge. Experiments from real cases show promising results, especially that our approach is able to distinguish more finely a skilled human from a novice, and that the latter would benefit from assistance regarding specific skills.